Server +

Study Guide

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Server +TM
Study Guide

Gary Govanus
with William Heldman
Jarret Buse

San Francisco Paris Dsseldorf Soest London

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Associate Publisher: Neil Edde

Contracts and Licensing Manager: Kristine OCallaghan
Acquisitions and Developmental Editor: Elizabeth Hurley
Editor: Linda Stephenson
Production Editors: Judith Hibbard, Shannon Murphy
Technical Editors: Scott Warmbrand, Donald Fuller
Contributor: Rebecca Monson
Book Designer: Bill Gibson
Graphic Illustrator: Tony Jonick
Electronic Publishing Specialist: Susie Hendrickson
Proofreaders: Jennifer Campbell, Nanette Duffy, Amey Garber, Leslie E. H. Light, Yariv Rabinovitch
Indexer: Nancy Guenther
CD Coordinator: Erica Yee
CD Technician: Kevin Ly
Cover Designer: Archer Design
Cover Photographer: Tony Stone Images
Copyright 2001 SYBEX Inc., 1151 Marina Village Parkway, Alameda, CA 94501. World rights reserved. No part of this
publication may be stored in a retrieval system, transmitted, or reproduced in any way, including but not limited to photocopy,
photograph, magnetic, or other record, without the prior agreement and written permission of the publisher.
Library of Congress Card Number: 00-109136
ISBN: 0-7821-2893-9
SYBEX and the SYBEX logo are either registered trademarks or trademarks of SYBEX Inc. in the United States and/or other
countries.
Screen reproductions produced with FullShot 99. FullShot 99 1991-1999 Inbit Incorporated. All rights reserved.
FullShot is a trademark of Inbit Incorporated.
The CD interface was created using Macromedia Director, COPYRIGHT 1994, 1997-1999 Macromedia Inc. For more
information on Macromedia and Macromedia Director, visit http://www.macromedia.com.
Internet screen shots using Microsoft Internet Explorer 5.5 reprinted by permission from Microsoft Corporation.
Sybex is an independent entity from CompTIA, and not affiliated with CompTIA in any manner. Neither CompTIA nor
Sybex warrants that use of this publication will ensure passing the relevant exam. Server+ is either a registered trademark
or trademark of CompTIA in the United States and/or other countries.
TRADEMARKS: SYBEX has attempted throughout this book to distinguish proprietary trademarks from descriptive terms
by following the capitalization style used by the manufacturer.
The author and publisher have made their best efforts to prepare this book, and the content is based upon final release
software whenever possible. Portions of the manuscript may be based upon pre-release versions supplied by software
manufacturer(s). The author and the publisher make no representation or warranties of any kind with regard to the completeness or accuracy of the contents herein and accept no liability of any kind including but not limited to performance,
merchantability, fitness for any particular purpose, or any losses or damages of any kind caused or alleged to be caused
directly or indirectly from this book.
Manufactured in the United States of America
10 9 8 7 6 5 4 3 2 1

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

To Our Valued Readers:

Sybex is proud to serve as a cornerstone member of the Server+ Advisory Committee. Just as CompTIA
is committed to establishing measurable standards for certifying individuals who will support computer
and networking systems in the future, Sybex is committed to providing those individuals with the skills
needed to meet those standards. By working alongside CompTIA, and in conjunction with other
esteemed members of the Server+ committee, it is our desire to help bridge the knowledge and skills gap
that currently confronts the IT industry.
Sybex expects the Server+ program to be well received, both by companies seeking qualified technical
staff and by the IT training community. Along with the existing line of vendor-neutral certifications
from CompTIA, including A+, Network+, and I-Net+, the Server+ certification should prove to be an
invaluable asset in the years ahead.
Our authors and editors have worked hard to ensure that the Server+ Study Guide is comprehensive, indepth, and pedagogically sound. Were confident that this book will meet and exceed the demanding
standards of the certification marketplace and help you, the Server+ exam candidate, succeed in your
endeavors.
Good luck in pursuit of your Server+ certification!

Neil Edde
Associate PublisherCertification
Sybex, Inc.

SYBEX Inc. 1151 Marina Village Parkway, Alameda, CA 94501

Tel: 510/523-8233
Fax: 510/523-2373 HTTP://www.sybex.com

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Software License Agreement: Terms and Conditions

The media and/or any online materials accompanying this book
that are available now or in the future contain programs and/or
text files (the "Software") to be used in connection with the book.
SYBEX hereby grants to you a license to use the Software, subject
to the terms that follow. Your purchase, acceptance, or use of the
Software will constitute your acceptance of such terms.
The Software compilation is the property of SYBEX unless otherwise indicated and is protected by copyright to SYBEX or other
copyright owner(s) as indicated in the media files (the
"Owner(s)"). You are hereby granted a single-user license to use
the Software for your personal, noncommercial use only. You
may not reproduce, sell, distribute, publish, circulate, or commercially exploit the Software, or any portion thereof, without the
written consent of SYBEX and the specific copyright owner(s) of
any component software included on this media.
In the event that the Software or components include specific
license requirements or end-user agreements, statements of condition, disclaimers, limitations or warranties ("End-User
License"), those End-User Licenses supersede the terms and conditions herein as to that particular Software component. Your
purchase, acceptance, or use of the Software will constitute your
acceptance of such End-User Licenses.
By purchase, use or acceptance of the Software you further agree
to comply with all export laws and regulations of the United
States as such laws and regulations may exist from time to time.
Software Support
Components of the supplemental Software and any offers
associated with them may be supported by the specific
Owner(s) of that material but they are not supported by
SYBEX. Information regarding any available support may
be obtained from the Owner(s) using the information provided in the appropriate read.me files or listed elsewhere on
the media.
Should the manufacturer(s) or other Owner(s) cease to offer support or decline to honor any offer, SYBEX bears no responsibility. This notice concerning support for the Software is provided
for your information only. SYBEX is not the agent or principal of
the Owner(s), and SYBEX is in no way responsible for providing
any support for the Software, nor is it liable or responsible for any
support provided, or not provided, by the Owner(s).
Warranty
SYBEX warrants the enclosed media to be free of physical defects
for a period of ninety (90) days after purchase. The Software is
not available from SYBEX in any other form or media than that
enclosed herein or posted to www.sybex.com. If you discover a
defect in the media during this warranty period, you may obtain
a replacement of identical format at no charge by sending the
defective media, postage prepaid, with proof of purchase to:

SYBEX Inc.
Customer Service Department
1151 Marina Village Parkway
Alameda, CA 94501
(510) 523-8233
Fax: (510) 523-2373
e-mail: [email protected]
WEB: HTTP://WWW.SYBEX.COM
After the 90-day period, you can obtain replacement
media of identical format by sending us the defective disk,
proof of purchase, and a check or money order for $10,
payable to SYBEX.
Disclaimer
SYBEX makes no warranty or representation, either expressed or
implied, with respect to the Software or its contents, quality, performance, merchantability, or fitness for a particular purpose. In
no event will SYBEX, its distributors, or dealers be liable to you
or any other party for direct, indirect, special, incidental, consequential, or other damages arising out of the use of or inability to
use the Software or its contents even if advised of the possibility of
such damage. In the event that the Software includes an online
update feature, SYBEX further disclaims any obligation to provide this feature for any specific duration other than the initial
posting.
The exclusion of implied warranties is not permitted by some
states. Therefore, the above exclusion may not apply to you.
This warranty provides you with specific legal rights; there may
be other rights that you may have that vary from state to state.
The pricing of the book with the Software by SYBEX reflects the
allocation of risk and limitations on liability contained in this
agreement of Terms and Conditions.
Shareware Distribution
This Software may contain various programs that are distributed
as shareware. Copyright laws apply to both shareware and ordinary commercial software, and the copyright Owner(s) retains all
rights. If you try a shareware program and continue using it, you
are expected to register it. Individual programs differ on details of
trial periods, registration, and payment. Please observe the
requirements stated in appropriate files.
Copy Protection
The Software in whole or in part may or may not be copy-protected or encrypted. However, in all cases, reselling or redistributing these files without authorization is expressly forbidden
except as specifically provided for by the Owner(s) therein.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

This book is dedicated to my two daughters, Denise Boedeker and Dawn

Carpenter. Thank you for letting me be a part of your life and giving me the
opportunity to watch you grow into such wonderful women.
Gary Govanus

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Acknowledgments
You know, this is the toughest part of the entire book to write. You may not
believe it, but it is true. So many people have done so much to get this book on
the shelf and into your hands that it is not possible for me to list them all. I also
have the task of trying to have you understand how important all of those people are to this project. Believe me, it is much easier explaining how Ethernet or
a router works than to try to explain the differences between an acquisition
and development editor, a production editor, and an editor!
Most importantly, I would like to thank my wife, Bobbi, for all her love
and understanding during this process. I have been writing for Sybex almost
continuously for two years, and she has been wonderful during the whole
time. It is a lot harder than it sounds, because when there are deadlines, or
I am trying to teach and to write at the same time, something has to give, and
usually it is the attention I pay to her. She knows how much I love doing this,
so she puts her needs on the back burner. She really is a wonderful woman,
and I am very lucky to have her in my life.
There are others who get shortchanged when I write. I sometimes have to
really work to find the time to make my daughters, Dawn and Denise, crazy.
Fortunately, they are now old enough where they have very full, successful
lives of their own. I dont get to see my grandkids nearly enough, so as soon
as I finish this thing, I am taking them to Disney World. Brandice has been
there before (several times), but CJ and Courtney havent so it will be a treat
for Poppy to see the wonder in their eyes. My parents have not had as much
of my time as they deserve either, and for that I am sorry. Finally, there is my
best friend, John Hartl. John is a quiet man, but can do a wonderful job of
laying guilt. He did it when he pointed out that seeing your best friend once
a year was not enough, and he was tired of me using the *&^% book as an
excuse. He is right!
Now for the people on the production team. This is the first book I have
done with Elizabeth Hurley. Since she approached me with this project, she
has been promoted to an acquisition and developmental editor, a position
she richly deserves. Soon she will be running the place. Her good humor and
infectious laugh always can brighten my day. I hope that this book will justify the faith she has had in me. Every time I came to her with a question, she
would say, Gary, you do what you think is best, I trust you completely.
You have no idea how close we came to changing this book into the novel I
always wanted to write!

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Judith Hibbard is the Production Editor on this project. She is an amazing

woman, because she actually fought to get on this book because she liked
working with me on a Windows 2000 book we did together. She is the stabilizing factor in the process, able to handle with dignity and grace anything
life or I can throw at her. Whenever I would call and rant and rave about
something, she would patiently listen, add insightful comments here and
there, and then bring me back to reality. I just love working with her, and the
wonderful part is that we have nurtured a friendship besides a working relationship. I cherish that.
Linda Stephenson is also truly a brave woman. This is book number three
that she has worked on with me, and she is the editor who has to read all this
stuff and try to make sense out of it. Linda does really write a lot of the book.
I do the research, I try and get the information on the page in some readable
format, but she completes the job. How she does all this with her busy life, I
will never know! She even volunteers to take busloads of children on weeklong
field trips.
Thanks also to my great friend, Becky Monson. Becky and I have been
talking about working together for years, and have never done it. She was
kind of drafted into writing some of the sample test questions, like a couple
of hundred. I am amazed that she is still talking to me! Becky, I told you that
wasnt a fun job!
I would be remiss if I did not also thank Bill Heldman and Jarret Buse. They
helped me make sure that this book provides the widest possible coverage of
CompTIAs Server+ objectives. Their analysis and contributions were invaluable to me.
Thanks also to my two tech editors, Scott Warmbrand and Donald Fuller.
They were the ones who had to read through all the material and straighten
me out when I went technically astray.
Thanks also to all those other people at Sybex that I dont know. These
include the graphic artist, Tony Jonick; the electronic publication specialist,
Susie Hendrickson; the proofreaders, Jennifer Campbell, Nanette Duffy,
Amey Garber, Leslie L. H. Light, and Yariv Rabinovitch; and the indexer,
Nancy Guenther.
A really big thanks to Senoria Bilbo-Brown. She is the one who made sure
the advance checks got to me in short order. Sen, you know I love ya!
And, of course, thanks to you the reader for buying this book and using
it to advance your career. May you not only pass the test, but smoke it!
Gary Govanus

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Introduction

he Server+ certification tests are sponsored by the Computing Technology Industry Association (CompTIA) and supported by several of the
computer industrys biggest vendors (for example, Compaq, IBM, and
Microsoft). This book was written to provide you with the knowledge you
need to pass the exam for Server+ certification. Server+ certification gives
employers a benchmark for evaluating their employees knowledge. When
an applicant for a job says, Im Server+ certified, the employer can be
assured that the applicant knows the fundamental server and networking
concepts. For example, a Server+ certified technician should know the difference between the various types of hard disk subsystems and how to configure them, the differences between various server types, and the
advantages and disadvantages of different network operating systems.
This book was written at an intermediate technical level; we assume that you
already know some of the information in the A+ certification and know about
hardware basics. The exam itself covers basic server topics as well as some more
advanced issues, and it covers some topics that anyone already working as a
technician, whether with computers or not, should be familiar with. The exam
is designed to test you on these topics in order to certify that you have enough
knowledge to intelligently discuss various aspects of server operations.
Weve included review questions at the end of each chapter to give you a
taste of what its like to take the exam. If youre already working as a network
administrator, we recommend you check out these questions first to gauge
your level of knowledge. You can begin measuring your level of expertise by
completing the assessment test at the end of this Introduction. Your score will
indicate which areas need improvement. You can use the book mainly to fill
in the gaps in your current knowledge of servers.
If you can answer 80 percent or more of the review questions correctly for
a given chapter, you can probably feel safe moving on to the next chapter. If
youre unable to answer that many correctly, reread the chapter and try the
questions again. Your score should improve.

Dont just study the questions and answersthe questions on the actual
exam will be different from the practice ones included in this book and on the
CD. The exam is designed to test your knowledge of a concept or objective, so
use this book to learn the objective behind the question.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

xxvi

Introduction

What Is Server+ Certification?

The Server+ certification program was developed by the Computer Technology Industry Association (CompTIA) to provide an industry-wide means of
certifying the competency of a Server Hardware Specialist. The Server+ certified diploma, which is granted to those who have attained the level of
knowledge and troubleshooting skills needed to provide capable support in
the field of networked servers, is similar to other certifications in the computer industry. For example, Novell offers the Certified Novell Engineer
(CNE) program to provide the same recognition for network professionals
who deal with its NetWare products, and Microsoft has its Microsoft Certified Service Engineer (MCSE) program. The theory behind these certifications is that if you need to have service, you would sooner call a technician
who has been certified in one of the appropriate certification programs than
the first so-called expert in the phone book.
What is a Server Hardware Specialist? According to CompTIA, a Server
Hardware Specialist is someone who spends time solving problems to ensure
that servers are functional and applications remain available. The specialist
should have an in-depth understanding of how to plan a network and how to
install, configure, and maintain a server. This should include knowing the hardware that goes into a server implementation, how data storage subsystems
work, the basics of data recovery, and how I/O subsystems work. In short, the
Server Hardware Specialist should have the following competencies:

Have a working knowledge of troubleshooting, physical security, and

disaster recovery. Be able to recover from a server failure.

Provide high availability of servers.

Understand hardware configuration and network connectivity.

Install and configure server hardware to meet the operating system

prerequisites.

Understand current and emerging data storage and data transfer

technologies.

Understand networking protocols.

Provide second-level support for resellers and end users.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Introduction

xxvii

Perform routine maintenance on server systems, data storage systems,

and other network devices

Plan and carry out an upgrade without impacting users.

CompTIA recommends that the candidate for Server+ Certification have

between 18 and 24 months experience in the server technology industry, as
well as some experience running a server. In addition, it is assumed that you
have at least one other IT certification, like the CompTIA A+, a Compaq
ACT, the Novell CNA, Microsoft MCP, HP Star, SCO, or Banyan.

Why Become Server+ Certified?

There are several good reasons to get your Server+ certification. The CompTIA Candidates Information packet lists five major benefits:

It demonstrates proof of professional achievement.

It increases your marketability.

It provides greater opportunity for advancement in your field.

It is increasingly found as a requirement for some types of advanced

training.

It raises customer confidence in you and your companys services.

Provides Proof of Professional Achievement

The Server+ certification is quickly becoming a status symbol in the computer service industry. Organizations with members in the computer service
industry are recognizing the benefits of Server+ certification and are pushing
for their members to become certified. And more people every day are putting the Server+ Certified emblem on their business cards.

Increases Your Marketability

Server+ certification makes individuals more marketable to potential
employers. Also, Server+ certified employees might receive a higher base
salary because employers wont have to spend as much money on vendorspecific training.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

xxviii

Introduction

Provides Opportunity for Advancement

Most raises and advancements are based on performance. Server+ certified
employees work faster and more efficiently, thus making them more productive. The more productive employees are, the more money they will make for
their company. And, of course, the more money they make for the company,
the more valuable they will be to the company. So if an employee is Server+
certified, her chances of getting promoted will be greater.

Fulfills Training Requirements

Server+ certification is recognized by most major computer hardware vendors, including (but not limited to) IBM, Hewlett-Packard, Apple, and Compaq. Some of these vendors will apply Server+ certification toward
prerequisites in their own respective certification programs.

Raises Customer Confidence

As the Server+ certified qualification becomes better known among computer owners, more of them will feel that the Server+ certified person is more
qualified to work on their computer equipment than a non-certified technician is.

How to Become Server+ Certified

Server+ certification is available to anyone who passes the tests. You dont have
to work for any particular company. Its not a secret society. It is, however, an
elite group. In order to become Server+ certified, you must pass the test.
The exam is administered by Prometric and can be taken at any Prometric Testing Center. If you pass the exam, you will get a certificate in the
mail from CompTIA saying that you have passed, and you will also receive
a lapel pin and business card. To find the Prometric training center nearest
you, call (800) 755-EXAM (755-3926).
To register for the tests, call Sylvan at (800) 77-MICRO (776-4276). Youll
be asked for your name, Social Security number (an optional number may be
assigned if you dont wish to provide your Social Security number), mailing
address, phone number, employer, when and where (i.e., which Prometric testing center) you want to take the test, and your credit card number. Arrangement
for payment must be made at the time of registration.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Introduction

xxix

It is possible to pass this test without any reference materials, but only if
you already have the knowledge and experience that come from reading
about and working with servers. Even experienced server people tend to
have what you might call a 20/80 situation with their computer knowledgethey may use 20 percent of their knowledge and skills 80 percent of
the time, and rely on manuals, guesswork, the Internet, or phone calls for the
rest. By covering all the topics that are tested by the exam, this book can help
you refresh your memory concerning topics that, until now, you seldom
used. (It can also serve to fill in gaps that, lets admit, you may have tried to
cover up for quite some time.) Further, by treating all the issues that the
exam covers, this book can serve as a general field guide, one that you may
want to keep with you as you go about your work.

In addition to reading the book, you might consider practicing these objectives
through an internship program. (After all, all theory and no practice make for a
poor technician.)

Who Should Buy This Book?

If you are one of the many people who want to pass the Server+ exam, and
pass it confidently, then you should buy this book and use it to study for the
exam. This book was written with one goal in mind: to prepare you for the
challenges of the real IT world, not just to pass the Server+ exam. This study
guide will do that by describing in detail the concepts on which youll be
tested.

How to Use This Book and CD

This book includes several features that will make studying for the Server+
exam easier. At the beginning of the book (right after this introduction, in
fact) is an assessment test that you can use to check your readiness for the
actual exam. Take the exam before you start reading the book. It will help
you determine the areas you need to brush up on. You can then focus on
these areas while reading the book. The answers to the assessment test
appear on a separate page after the last question of the test. Each answer also
includes an explanation and a note telling you in which chapter this material
appears.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

xxx

Introduction

To test your knowledge as you progress through the book, check out the
review questions at the end of each chapter. As you finish each chapter,
answer the review questions and then check to see if your answers are right
the correct answers appear on the page following the last review question.
You can go back to reread the section that deals with each question you got
wrong to ensure that you get the answer correctly the next time you are
tested on the material.
On the CD-ROM youll find two sample exams. You should test your
knowledge by taking the practice exam when you have completed the book
and feel you are ready for the Server+ exams. Take this practice exam just as
if you were actually taking the Server+ exam (i.e., without any reference
material). When you have finished the practice exam, move on to the bonus
exam to solidify your test-taking skills. If you get more than 90 percent of the
answers correct, youre ready to go ahead and take the real exam.
The CD-ROM also includes several extras you can use to bolster your
exam readiness:
Electronic flashcards You can use these 150 flashcard-style questions to
review your knowledge of Server+ concepts. They are available for PCs
and handheld devices. You can download the questions right into your
Palm device for quick and convenient reviewing anytime, anywhere
without your PC!
Test engine The CD-ROM includes all of the questions that appear in this
book: the assessment questions at the end of this introduction and all of the
chapter review questions. Additionally, it includes a practice exam and a
bonus exam. The questions appear much like they did in the book, but you
can also choose to randomize them. The randomized test will allow you to
pick a certain number of questions to be tested on, and it will simulate the
actual exam. Combined, these test engine elements will allow you to test your
readiness for the real Server+ exam.
Full text of the book in PDF If you are going to travel but still need to
study for the Server+ exam and you have a laptop with a CD-ROM drive, you
can take this entire book with you just by taking the CD-ROM. This book is
in Adobe Acrobat PDF format so it can be easily read on any computer.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Introduction

xxxi

Sybex as Server+ Cornerstone Members

Sybex is proud to serve as a Cornerstone Member of the Server+ Advisory
Committee. Just as CompTIA is committed to establishing measurable standards for certifying individuals who will support computer and networking
systems in the future, Sybex is committed to providing those individuals with
the skills needed to meet those standards. Attaining a Server+ certification is
evidence that the candidate possesses core competencies for vendor-neutral
server integration, management, and service. Server+ is the logical companion
to both the A+ and Network+ certifications. CompTIAs addition of Server+
to their current line of exams creates a true baseline knowledge across the
industry for those seeking to increase their IT skills. Sybex is committed to
bringing that information to you, the future Server+ certified administrator. By
working alongside CompTIA, and in conjunction with other esteemed members of the Server+ exam committee, it is our desire to help bridge the gap in
knowledge and skills that currently confronts the IT industry.
Sybex expects the Server+ program to be well received, both by companies
seeking qualified technical staff and by the IT training community. Along with
the existing line of vendor-neutral certifications from CompTIA, the Server+
certification should prove to be an invaluable asset in the years ahead.

The Exam Objectives

Behind every computer industry exam you can be sure to find exam objectivesthe broad topics in which the exam developers want to ensure your
competency. The official CompTIA Server+ exam objectives are listed here.

Server+ Exam Blueprint

The table that follows lists the areas (or in CompTIA terms, the domains)
measured by this examination and the approximate extent to which they
are represented.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

xxxii

Introduction

Job Dimension

% of Exam (approximate)

1.0: Installation

17%

2.0: Configuration

18%

3.0: Upgrading

12%

4.0: Proactive Maintenance

9%

5.0: Environment

5%

6.0: Troubleshooting and Problem

Determination

27%

7.0: Disaster Recovery

12%

1.0: Installation
1.1 Conduct pre-installation planning activities:

Plan the installation.

Verify the installation plan.

Verify hardware compatibility with operating system.

Verify power sources, space, UPS, and network availability.

Verify that all correct components and cables have been delivered.

1.2 Install hardware using ESD best practices (boards, drives, processors,

memory, internal cable, etc.):

Mount the rack installation.

Cut and crimp network cabling.

Install UPS.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Introduction

xxxiii

Verify SCSI ID configuration and termination.

Install external devices (e.g., keyboards, monitors, subsystems,

modem rack, etc.).

Verify power-on via power-on sequence.

2.0: Configuration
2.1 Check/upgrade BIOS/firmware levels (system board, RAID, control-

ler, hard drive, etc.).

2.2 Configure RAID.
2.3 Install NOS.

Configure network and verify network connectivity.

Verify network connectivity.

2.4 Configure external peripherals (UPS, external drive subsystems, etc.).

2.5 Install NOS updates to design specifications.
2.6 Update manufacturer-specific drivers.
2.7 Install service tools (SNMP, backup software, system monitoring

agents, event logs, etc.).

2.8 Perform Server baseline.
2.9 Document the configuration.

3.0: Upgrading
3.1 Perform full backup:

Verify backup.

3.2 Add Processors:

On single-processor upgrade, verify compatibility.

Verify N 1 stepping.

Verify speed and cache matching.

Perform BIOS upgrade.

Perform OS upgrade to support multiprocessors.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

xxxiv

Introduction

Perform upgrade checklist, including: locate/obtain latest test

drivers, OS updates, software, etc.; review FAQs, instructions,
facts and issues; test and pilot; schedule downtime; implement
ESD best practices; confirm that upgrade has been recognized;
review and baseline; document upgrade.

3.3 Add hard drives:

Verify that drives are the appropriate type.

Confirm termination and cabling.

For ATA/IDE drives, confirm cabling, master/slave, and potential

cross-brand compatibility.

Upgrade mass storage.

Add drives to array.

Replace existing drives.

Integrate into storage solution and make it available to the operating

system.

Perform upgrade checklist, including: locate and obtain latest test

drivers, OS updates, software, etc.; review FAQs, instructions,
facts and issues; test and pilot; schedule downtime; implement
using ESD best practices; confirm that the upgrade has been
recognized; review and baseline; document the upgrade.

3.4 Increase memory:

Verify hardware and OS support for capacity increase.

Verify memory is on hardware/vendor compatibility list.

Verify memory compatibility (e.g., speed, brand, capacity, EDO,

ECC/non-ECC, SDRAM/RDRAM).

Perform upgrade checklist, including: locate and obtain latest test

drivers, OS updates, software, etc.; review FAQs, instructions,
facts and issues; test and pilot; schedule downtime; implement
using ESD best practices; confirm that the upgrade has been recognized; review and baseline; document the upgrade.

Verify that server and OS recognize the added memory

Perform server optimization to make use of additional RAM.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Introduction

xxxv

3.5 Upgrade BIOS/firmware:

Perform upgrade checklist, including: locate and obtain latest test

drivers, OS updates, software, etc.; review FAQs, instructions, facts
and issues; test and pilot; schedule downtime; implement using ESD
best practices; confirm that the upgrade has been recognized;
review and baseline; document the upgrade.

3.6 Upgrade adapters (e.g., NICs, SCSI cards, RAID, etc.):

Perform upgrade checklist, including: locate and obtain latest test

drivers, OS updates, software, etc.; review FAQs, instructions,
facts and issues; test and pilot; schedule downtime; implement
using ESD best practices; confirm that the upgrade has been recognized; review and baseline; document the upgrade.

3.7 Upgrade peripheral devices, internal and external:

Verify appropriate system resources (e.g., expansion slots, IRQ,

DMA, etc.).

Perform upgrade checklist, including: locate and obtain latest test

drivers, OS updates, software, etc.; review FAQs, instructions, facts
and issues; test and pilot; schedule downtime; implement using ESD
best practices; confirm that the upgrade has been recognized; review
and baseline; document the upgrade.

3.8 Upgrade system monitoring agents:

Perform upgrade checklist, including: locate and obtain latest test

drivers, OS updates, software, etc.; review FAQs, instructions,
facts and issues; test and pilot; schedule downtime; implement
using ESD best practices; confirm that the upgrade has been recognized; review and baseline; document the upgrade.

3.9 Upgrade service tools (e.g., diagnostic tools, EISA configuration,

diagnostic partition, SSU, etc.):

Perform upgrade checklist, including: locate and obtain latest test

drivers, OS updates, software, etc.; review FAQs, instructions, facts
and issues; test and pilot; schedule downtime; implement using ESD
best practices; confirm that the upgrade has been recognized; review
and baseline; document the upgrade.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

xxxvi

Introduction

3.10 Upgrade UPS:

Perform upgrade checklist, including: locate and obtain latest test

drivers, OS updates, software, etc.; review FAQs, instructions, facts
and issues; test and pilot; schedule downtime; implement using ESD
best practices; confirm that the upgrade has been recognized; review
and baseline; document the upgrade.

4.0: Proactive Maintenance

4.1 Perform regular backup.
4.2 Create baseline and compare performance.
4.3 Set SNMP thresholds.
4.4 Perform physical housekeeping.
4.5 Perform hardware verification.
4.6 Establish remote notification.

5.0: Environment
5.1 Recognize and report on physical security issues:

Limit access to server room and backup tapes.

Ensure physical locks exist on doors.

Establish anti-theft devices for hardware (lock server racks).

5.2 Recognize and report on server room environmental issues (temper-

ature, humidity/ESD/power surges, backup generator/fire suppression/flood considerations).

6.0: Troubleshooting and Problem Determination

6.1 Perform problem determination:

Use questioning techniques to determine what, how, when.

Identify contact(s) responsible for problem resolution.

Use senses to observe problem (e.g., smell of smoke, observation

of unhooked cable, etc.).

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Introduction

xxxvii

6.2 Use diagnostic hardware and software tools and utilities:

Identify common diagnostic tools across the following operating

systems: Microsoft Windows NT/2000, Novell Netware, UNIX,
Linux, and IBM OS/2.

Perform shutdown across the following operating systems:

Microsoft Windows NT/2000, Novell Netware, UNIX, Linux,
and IBM OS/2.

Select the appropriate tool.

Use the selected tool effectively.

Replace defective hardware components as appropriate.

Identify defective FRUs and replace with correct part.

Interpret error logs, operating system errors, health logs, and critical
events.

Use documentation from previous technician successfully.

Locate and effectively use hot tips (e.g., fixes, OS updates, E-support,
Web pages, CDs).

Gather resources to get problem solved.

Identify situations requiring call for assistance.

Acquire appropriate documentation.

Describe how to perform remote troubleshooting for a wake-on-LAN.

Describe how to perform remote troubleshooting for a remote alert.

6.3 Identify bottlenecks (e.g., processor, bus transfer, I/O, disk I/O, network

I/O, memory).
6.4 Identify and correct misconfigurations and/or upgrades.
6.5 Determine if problem is hardware, software, or virus related.

7.0: Disaster Recovery

7.1 Plan for disaster recovery:

Plan for redundancy (e.g., hard drives, power supplies, fans,

NICs, processors, UPS).

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

xxxviii

Introduction

Use the technique of hot swap, warm swap and hot spare to
ensure availability.

Use the concepts of fault tolerance/fault recovery to create a disaster recovery plan.

Develop disaster recovery plan.

Identify types of backup hardware.

Identify types of backup and restoration schemes.

Confirm and use off site storage for backup.

Document and test disaster recovery plan regularly, and update as

needed.

7.2 Restoring.

Identify hardware replacements.

Identify hot and cold sites.

Implement disaster recovery plan.

Tips for Taking the Server+ Exam

Here are some general tips for taking your exam successfully:

Bring two forms of ID with you. One must be a photo ID, such as a
drivers license. The other can be a major credit card or a passport.
Both forms must have a signature.

Arrive early at the exam center so you can relax and review your study
materials, particularly tables and lists of exam-related information.

Read the questions carefully. Dont be tempted to jump to an early

conclusion. Make sure you know exactly what the question is asking.

Dont leave any unanswered questions. Unanswered questions are

scored against you.

There will be questions with multiple correct responses. When there is

more than one correct answer, a message at the bottom of the screen
will prompt you to Choose all that apply. Be sure to read the messages displayed.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Introduction

xxxix

When answering multiple-choice questions youre not sure about, use

a process of elimination to get rid of the obviously incorrect options
first. This will improve your odds if you need to make an educated
guess.

On form-based tests, because the hard questions will eat up the most time,
save them for last. You can move forward and backward through the
exam. When the exam becomes adaptive, this tip will not work.

For the latest pricing on the exams and updates to the registration procedures, call Prometric at (800) 755-EXAM (755-3926) or (800) 77-MICRO
(776-4276). If you have further questions about the scope of the exams or
related CompTIA programs, refer to the CompTIA site at www.comptia.org/.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Assessment Test
1. In a Fibre Channel configuration, what constitutes a point-to-point link?
A. Arbitrated loop
B. Fabric
C. A bidirectional link that connects the N_ports on two nodes
D. Two NL_ports connected to two FL_ports
2. What do you call a list of IP addresses that can be assigned by an auto-

matic server process?

A. A DNS scope
B. A DHCP scope
C. FTP
D. Lost
3. You have just purchased a mainboard that supports dual processors.

Which Pentium III processors can be used on the board?

A. Any Xeon with any P-II
B. P-IIIs of the same speed
C. Any P-III
D. Any P-II with any P-III
4. If you have a RAID 3 system made up of four 20GB drives, how much

usable disk storage space would you have?

A. 80GB
B. 60GB
C. 40GB
D. 20GB
E. Not enough information to make a determination.
5. The performance of a RISC processor depends on which of the following?
A. The code it is executing
B. The speed of the network card
C. The amount of RAM
D. The video card

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

xli

Assessment Test

6. Memory Interleaving is another way of doing which of the following:

A. Error checking
B. Accessing information stored on the memory chip
C. Determining parity
D. Installing chips
7. What happens when an ECC memory module determines that corruption

has occurred in 1 bit?

A. The problem is immediately corrected and the end user is none the

wiser.
B. An error message pops up on the screen describing the error to the
end user and giving the user a chance to fix the problem.
C. An entry is made in the memory error log, but the system continues
to operate.
D. The system is halted.
8. What happens when a parity-checking memory module determines

that corruption has occurred?

A. The problem is immediately corrected and the end user is none

the wiser.
B. An error message pops up on the screen describing the error to the

end user and giving the user a chance to fix the problem.
C. An entry is made in the memory error log, but the system continues
to operate.
D. The system is halted.
9. How many interrupts are available with PCI?
A. 64
B. 32
C. 16
D. 8
10. How can you configure load balancing in a PCI Bridged environment?
A. Configure one bridge as a master, and the other as a slave.
B. You will have to buy special devices to make this work.
C. You will have to purchase a special connector.
D. Load balancing is not recommended in bridged environment.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Assessment Test

11. What are the three software layers of I2O?

A. OS/2
B. OSM
C. NLM
D. IML
E. CDM
F. HDM
12. What are three possible configurations for an ATA/IDE device?
A. Master, with slave present
B. Slave, with a master present
C. Slave, no master present
D. Master, no slave present
E. Leader
F. Follower
13. What is the bandwidth of PCI?
A. 133 MBytes/second
B. 156 MBytes/second
C. 64 MBytes/second
D. 32 MBytes/second
14. What is the N_ports unique address called?
A. N_port Identifier
B. Well-known port address
C. IP address
D. VPN address
15. When cabling a building, what should you do?
A. Only use fiber optic cable.
B. Always use copper conduit.
C. Always check local building codes.
D. Assume that you do not need a permit.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

xlii

xliii

Assessment Test

16. Four network cards grouped together for Load Balancing will have

how many IP addresses?

A. Four
B. Three
C. Two
D. One
17. ATA 100 can also be referred to as which of the following?
A. ATA Parallel
B. ATA Serial
C. ATA
D. ATA Bipolar
18. Why would you install a bridge?
A. To route packets
B. To minimize traffic on a network segment
C. For security purposes
D. To dynamically assign IP Addresses
19. You have a single network card with four ports on it. What can that

card not be configured to do?

A. Adapter Load Balancing
B. Adapter Teaming
C. Adapter Fault Tolerance
20. You have a very important database on your network. You decide to

check to make sure the database is getting backed up, so you try to
restore one of the files to another server. You find the file was not
backed up. What is a likely reason for this happening?
A. The file had not been accessed that day.
B. The tape backup program cannot back up open files.
C. The tape backup program cannot back up files that big.
D. The tape backup program did not run.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Assessment Test

xliv

21. In every SCSI-3 bus, how many terminators are there?

A. Four
B. Three
C. Two
D. One
E. One per device
22. Which type of application server would be used by gamers?
A. Peer to peer
B. Distributed
C. Dedicated
D. None of the above
23. Pick the levels of cache that can be present in a computer with a Pentium

III Gigahertz processor.

A. L1
B. L2
C. L3
D. L4
24. If a DHCP Server is not on the same subnet as the hosts it serves, what

must be configured?
A. A DNS Server
B. A relay Agent
C. Another DHCP Server
D. SMTP
E. DMI
25. Name three ways NICs can work together.
A. Adapter Grouping
B. Adapter Fault Tolerance
C. Adapter Virtual Private Networks
D. Adapter Load Balancing
E. Adapter Teaming

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

xlv

Assessment Test

26. What does PIO stand for?

A. Progressive Input/Output
B. Processor Input/Output
C. Programmable Input/Output
D. Programmed Input/Output
27. If a server is rated at 4U, what does that mean?
A. It is a rack-mounted server that will use 4 milliamps of power.
B. It is a rack-mounted server that will use 400 volts of power.
C. It is a rack-mounted server that can fit in a 40-inch rack.
D. It is a rack-mounted server that will cover four of the mounting

holes in the rack.

28. What type of server resolves DNS names to IP addresses?
A. DHCP
B. DNS
C. UDP
D. SMTP
29. Why are maintenance logs important?
A. They provide a clear picture of what the service techs have been

doing.
B. They provide a background of what has been done to a computer.
C. They provide an instruction manual for doing routine tasks.
30. What is the plenum?
A. The type of metallic shielding surrounding a fiber optic cable
B. The type of cable used in fiber optic installations
C. The air space between the ceiling and the actual roof of a building
D. Precious metal like gold
31. With which Internet standard protocol is Active Directory accessed?
A. SNMP
B. SMTP
C. LDAP
D. POP3

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Assessment Test

xlvi

32. Will an ATA 100 device use the same type cable as an ATA 66 device?
A. No
B. Yes
33. A BNC connector is used on what type of Ethernet implementation?
A. ThinNet
B. Thicknet
C. UTP
D. STP
34. How many terminators are there on a ThinNet network?
A. One
B. Two
C. One for every 50 hosts
D. One for every 100 hosts
35. Which is true of fiber optics?
A. It is affected by EMI.
B. It is affected by heat.
C. The cable can be made of glass.
D. The cable is always made of copper.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

xlvii

Answers to Assessment Test

Answers to Assessment Test

1. C. In Fibre Channel, a point-to-point connection is a bidirectional link

that connects the N_ports on two nodes.

2. B. It is called a DHCP scope.
3. B. With Pentium III processors, the multiplier and the FSB must

match.
4. B. You would have three 20GB drives for storage and one 20GB drive

for parity. Therefore, you would have 60GB of usable storage space.
5. A. The performance of the RISC processor depends on the code it is

executing.
6. B. Memory interleaving is a way of quickly getting access to informa-

tion stored on the memory chip.

7. A. With an ECC memory module, if there is a problem with 1 bit, it

will be detected and corrected. ECC memory can determine corruption of up to 4 bits, but with anything over 1 bit, the system is halted.
8. D. With parity, if it is determined there has been some corruption, the

system is halted.
9. C. PCI can use up to 16 interrupts.
10. D. If you are using a bridged architecture, load balancing is not rec-

ommended.
11. B, D, F. I2O is made up of three software layers: the OS Services Mod-

ule (OSM), the I2O Messaging Layer (IML), and the Hardware Device
Module (HDM).
12. A, B, D. IDE devices can be a master with no slave present, a master

with a slave present, and a slave with a master present.

13. A. The PCI bandwidth is 133 MBytes/second.
14. A. The N_port has a unique address, called the N_port Identifier
15. C. When cabling a building, you should check the local building

codes. These codes will vary by locality.

16. One. A group of network cards used in Load Balancing will have one

IP address.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Answers to Assessment Test

xlviii

17. B. ATA 100 can also be referred to as ATA Serial.

18. B. A bridge would be used to minimize traffic between network

segments.
19. C. Fault Tolerance requires at least two cards, not at least two ports.
20. B. Many tape backup programs are not capable of backing up open

database files without an add-in component.

21. C. There will always be two terminators, one at each end of the SCSI

chain.
22. A. A peer to peer application server would be the type that may be

used by gamers.
23. A, B. There are only two types of cache, L1 and L2.
24. B. A relay Agent must be configured.
25. B, D, E. Adapters can work together with Load Balancing, Fault Tol-

erance, or Teaming.
26. D. PIO is the abbreviation for Programmed Input/Output.
27. D. The U rating is the number of mounting holes that the device will

utilize. A U is 4.445 centimeters or 1.75 inches.

28. B. A DNS Server resolves a DNS name to an IP address.
29. B. Maintenance logs provide a background of what has been done to

a computer.
30. C. The plenum is the space between the drop down ceiling and the

roof where you can run cables.

31. C. Active Directory is based on LDAP.
32. A. No, with ATA 100 the cabling changes.
33. A. BNC Connectors are used on ThinNet Ethernet networks.
34. A. There is one and only one terminator on a ThinNet network.
35. C. Fiber cable is made of glass or plastic.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Chapter

Disk Subsystems
SERVER+ EXAM OBJECTIVES COVERED IN
THIS CHAPTER:

1.2 Install hardware using ESD best practices (boards, drives,
processors, memory, internal cable, etc.).


Mount the rack installation.

Cut and crimp network cabling.

Install UPS.

Verify SCSI ID configuration and termination.

Install external devices (e.g., keyboards, monitors,

subsystems, modem rack, etc.).

Verify power-on via power-on sequence.

2.2 Configure RAID.

3.3 Add hard drives.


Verify that drives are the appropriate type.

Confirm termination and cabling.

For ATA/IDE drives, confirm cabling, master/slave, and

potential cross-brand compatibility.

Upgrade mass storage.

Add drives to array.

Replace existing drives.

Integrate into storage solution and make it available to the

operating system.

Perform upgrade checklist, including: locate and obtain

latest test drivers, OS updates, software, etc.; review FAQs,
instructions, facts and issues; test and pilot; schedule
downtime; implement using ESD best practices; confirm that
the upgrade has been recognized; review and baseline;
document the upgrade.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com


3.6 Upgrade adapters (e.g., NICs, SCSI cards, RAID, etc.).



Perform upgrade checklist, including: locate and obtain

latest test drivers, OS updates, software, etc.; review FAQs,
instructions, facts and issues; test and pilot; schedule
downtime; implement using ESD best practices; confirm that
the upgrade has been recognized; review and baseline;
document the upgrade.

7.1 Plan for disaster recovery.



Plan for redundancy (e.g., hard drives, power supplies, fans,

NICs, processors, UPS).

Use the technique of hot swap, warm swap, and hot spare to
ensure availability.

Use the concepts of fault tolerance/fault recovery to create a

disaster recovery plan.

Develop disaster recovery plan.

Identify types of backup hardware.

Identify types of backup and restoration schemes.

Confirm and use off site storage for backup.

Document and test disaster recovery plan regularly, and

update as needed.

7.2 Restoring


Identify hardware replacements.

Identify hot and cold sites.

Implement disaster recovery plan.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

ont you just hate it? You buy the darn book, hoping to be
slowly and gently eased into the studying process, and in the very first chapter
the author nails you with a ton of objectives. Well, take heart, if you have
passed the Network+ test and the A+ test, about 20% of the material in this
book will be old hat! You will already know it.
The Server+ exam is designed to give you background into the inner
workings of your local server platform. Throughout this book we are going
to be talking about the different types of hardware that make up a server, the
different types of servers that can be put into your network, the different
types of server operating systems, how to care for your servers, and how to
fix them if they break. I suppose we could subtitle this book, The Care and
Feeding of Network Servers.
To make this daunting task a little easier, we are going to break it down
into chunks. As you can see, the first chunk deals with the disk subsystem of
the server. In this first chapter, we will cover four basic areas: Logical and
Physical Drives, SCSI, RAID, and hot swappable. Well spend some time
hashing out terminology, discussing strengths and weaknesses, and looking
at fault tolerances. So lets get to it.

For complete coverage of objective 1.2, please also see Chapters 6, 7, and 8. For
complete coverage of objective 3.3, please also see Chapter 2. For complete coverage of objective 3.6, please also see Chapters 2 and 10. For complete coverage
of objectives 7.1 and 7.2, please also see Chapter 12.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Chapter 1

Disk Subsystems

Its All About Perception

In this section, we are going to talk about Logical Drives and Physical
Drives and describe their functionality. We will take a look at how the people
who use the server view the disk subsystem.
Now, if your users are like all the users I have ever dealt with, 95% of
them dont care how or why the network operatesthey just want it to work
every time. Does that sound familiar? When it gets into the subject of drives,
they could care less, as long as they can store and retrieve their information.
And that is just the way it should be.

Logical Drives
Every network that I have ever worked on has had a drive letter mapped to
an area that was fondly referred to as the Users home directory. Depending
on the network operating system you are using, it may be called the Users
directory, or the mount point, or something else esoteric, but every system
has one. This is the place where your users can store their highly personal,
private user stuff. You know, like all the jokes they received via e-mail over
the weekend. Anyway, to make this drive easier for users to access, it is
assigned a drive letter; for instance, in my network it is the H: drive. Now,
users dont refer to this area as their Users home directorythey simply call
it their H: drive. Well, what exactly is their H: drive?
If you were to ask an end user that question, he would probably tell you
that somewhere, back in the deep dark reaches of the computer room, there
would be a wall. Mounted on this wall would be dozens of physical hard
drives and each one of these hard drives would have a little plastic strip on
it, with a name. So, if you wanted to find Elizabeths H: drive, you would
simply find the wall, look to the strips that handled the Es, and there, about
halfway down, would be the drive for Elizabeth. In reality, the Users home
directory is just thatit is a directory that is part of a much larger directory
structure. Using Microsoft Explorer, Figure 1.1 is a small sample of a Users
directory.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Its All About Perception

FIGURE 1.1

A sample of the Users directory

Now, I suppose right from the start we should get the biases out in the open.
First of all, I have been working with computers since the 80s, back when DOS
was king and GUI was something slimy. So, I have some problems that you
should know about. The first is with the interchangeable term directories and
folders. In the world of GUI, a folder represents a storage area created on a
disk for the storage and retrieval of information. In the world of DOS, the same
thing was called a directory. Being an old dog, I find it hard to learn new tricks,
so if you see the word directory, and you are more comfortable with folder, go
for it. In my world, in this context, they are interchangeable.

Depending on the network operating system you are using, these areas are
referred to as mapped drives, shares, or mount points. It all amounts to the
same thing. A drive letter has been assigned as a pointer to a particular directory or folder on a bigger physical device. It does not even have to be a network operating system. DOS will support up to 23 Logical Drives on a

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Chapter 1

Disk Subsystems

system. As far as the user is concerned, it is a drive just like the C: drive in
their computer. As far as you are concerned, it is a Logical Drive, or drive letter that has been assigned as a pointer to a distinct directory or folder on a
larger physical device.

Physical Drives
If Logical Drives are pointers to directories or folders on physical devices, it
makes sense that the Physical Drive is what you can hold on to and install
into a file server. A Physical Drive can be a hard drive, a floppy drive, or even
removable storage.
These drives come in a variety of sizes, from the standard 1.44MB floppy
drive to the hard drives that can go over 30 gigabytes. The hard drives also
come in a variety of different technologies and configurations, with a wide
variety of different acronyms that you are going to have to be familiar with
things like IDE, EIDE, ATA, SCSI, RAID, hot swap, hot plug, and even hot
spare. Over the rest of this chapter and the next chapter, we will be talking
about the differences between these.

Real World Scenario

These little ditties will be thrown in throughout the book just to make sure that
we keep both feet on the ground and realize that there really is a difference
between the CompTIA testing world and the real world you work in every day.
Hopefully, the Real World Scenarios will help bring home some of the points
we discussed, as well as solidify how the information can be used.
The best way to track the difference between a logical drive and a physical
drive is to think of this common scenario: Many times I have had an
administrative assistant call me because he wanted some special access
to his bosss home directory. The problem was, the administrative assistant didnt know that it was his bosss home directory he wanted access
tohe thought it was the H: drive! When I would grant him rights to the
bosss home directory, I would map a drive for the administrative assistant, and give it a different drive letter, like the I: drive. The discussion
would then ensue that the assistant did not want access to the I: drivehe
needed access to the bosss H: drive. The whole thing could get really ugly.
Remember that a physical drive can contain hundreds of logical drives, but
a logical drive usually resides on just one physical drive.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

The World of SCSI

The World of SCSI

There are several limitations to the way EIDE handles devices. The major
limitations are the number of devices that can be controlled from a single paddle
card and the lack of redundancy. If one device fails, your entire subsystem is
destroyed. Small Computer System Interface (SCSI), on the other hand,
addresses those issues. SCSI is just another type of interface that is much more
extensible than IDE. Besides hard drives, SCSI will work with CD-ROMs and all
sorts of other wonderful things. Even from its early days, SCSI (pronounced
scuzzy) allowed you to have an unlimited number of devices strung together in
what is referred to as a daisy chain. SCSI was originally designed as a high-speed
system-level parallel interface. SCSI has evolved, and now there are all sorts of
different levels and speeds. We will explore each of them in this section.
Lets start with the definitions of the interfaces and how they are used.

SCSI-1 Narrow and Regular

One of the best things about SCSI is that SCSI devices are very flexible. They
can be installed in just about anythingMacs, PCs, SUN workstationsjust
about anything. It there is a SCSI controller made for it, the SCSI peripheral
will work with it. Also with SCSI, there is an intelligent controller. This SCSI
controller can be an installed interface board, or it can be built in to the
mainboard, just like the IDE paddle card. The difference is that the SCSI controller actually does make a difference in the way it controls devices. SCSI is
defined as an intelligent peripheral interface that makes use of high-level
communication between devices like hard drives or tape drives. Communications occur between an initiator (normally the computer) and a target
(usually the peripheral). Data transfer can occur in an asynchronous mode,
meaning it is not clocked, or in synchronous (clocked) mode.

When we start talking about synchronous and asynchronous, the term clocked
comes into play. If you are not familiar with what it means, think of your dial-up
connection to the Internet. In this case, you are using an asynchronous modem.
Communication occurs randomly, when and where you want it to. Since there is
no regularity, this is not clocked. With synchronous, whenever a communication
link is established, certain tasks are carried out at regularly timed intervals, and
therefore they are clocked.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Chapter 1

Disk Subsystems

Any messages and commands are transferred asynchronously. The data

bus in SCSI-1 was 8-bit, so it was considered SCSI Narrow.
The command process went something like this. Take a look at Figure 1.2.
FIGURE 1.2

SCSI chain for communication explanation

SCSI controller

Device 1

Device 2

Device 3

Device 4

As you can see, we have a SCSI controller and four SCSI devices. These
devices could be four hard drives or three hard drives and a tape drive, or
two hard drives, a tape drive, and a CD-ROM. You get the point. Anyway,
suppose the computer sent messages to the controller to write information to
Device 1 and to Device 3. With SCSI, the controller would send a signal to
Device 1 telling it that the controller had some work for it to do. Device 1
would respond, and the controller would send the information. Device 1
would send back an acknowledgement and the controller would now go on
to the information that had to be written to Device 3. The trick here is that
the controller would do things one step at time, and could not multitask.
That was changed in SCSI-2.
So, how does the controller know which device is which? Well, just like in
the diagram, each device is assigned its own unique device number, called the
SCSI address. The entire SCSI subsystem is referred to as the SCSI bus. The
SCSI address was configured in a variety of ways, including jumpers or rocker
switches. That way, when the controller needed to send information to that
device, it just used the appropriately addressed wire. In order to keep the signals on the wire, each SCSI bus had to be terminated at both ends. We will talk
more about termination after we get through defining the types of SCSI.
Not only is SCSI flexible in the kinds of computers it can work in, SCSI
is also flexible in the kinds of devices it can work with. For example, SCSI
can work with tape drives, hard drives, and CD-ROMs, to name a few.
These devices can be internal to the computer or external, in a separate case.
If the devices were internal, they would use a 50-pin ribbon cable. If the
devices were external, they would use a very thick, shielded cable that had a
Centronics 50-pin adapter on one end and a DB-25 connector on the other.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

The World of SCSI

SCSI was defined in 1986 and is a published American National Standards

Institute (ANSI) standard called SCSI-1 (X3.1311986).

SCSI-2 Wide and Fast

As SCSI technology began to mature, it received another ANSI standard
(X3.1311994). This was truly an upgrade from the previous version,
because it featured faster data transfer rates, and it also mandated the
structure of messages and commands to help improve compatibility.
With SCSI-2, communication shifted to synchronous data transfer, and the
width of the bus grew to 16 bits. The data transfer rate went from 2.5 to 10
Mbytes/second on an 8-bit data bus and from 5.0 to 20.0 Mbytes/second for
a 16-bit data bus. This change also necessitated changes to the connectors,
using a higher density connector, as shown in Figure 1.3.
FIGURE 1.3

A SCSI-2 connector

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

10

Chapter 1

Disk Subsystems

Now just to confuse you, when they widened the data transfer path from
8 to 16 bits, the 16-bit version of SCSI-2 was also referred to as Wide SCSI.
With the wider bus, the transfer speed climbed to 10.0 Mbytes/second; this
was referred to as Fast SCSI. With Fast SCSI, there was also some new terminologyinstead of Mbytes/second, there is Mega Transfer (MT). The MT
is a unit of measurement that refers to the rate of signals on the interface,
regardless of the width of the bus. So, as an example, if you have a 10MT
rate on a Narrow SCSI bus, the transfer rate would be 10 Mbytes/second. If,
however, you had the same 10MT rate on a Wide bus, it would result in a
20 Mbyte/second transfer rate. The developers finally took the Wide SCSI
technology and combined it with Fast SCSI, and that became Fast-Wide
SCSI, with a transfer speed of 40 Mbytes/second.
SCSI-2 was backwardly compatible with SCSI, but for maximum benefit, it
was suggested that you stick with one technology or the other, preferably
using a SCSI-2 controller with SCSI-2 devices. With both SCSI-1 and SCSI-2,
the number of peripherals that could be connected to any controller was seven.

SCSI-3
Although SCSI is maturing, it is not completely there yet. There are still some
limitations, like having no more than seven devices connected to any controller.
The next generation of SCSI, SCSI-3, takes care of some of that.
Now SCSI-3 is still a proposed ANSI standard, but there are a lot of devices
out there purporting to be SCSI-3. That is because the SCSI-3 documentation
took the very large SCSI-2 specifications (in excess of 400 pages) and split it
into smaller bite size chunks. These smaller documents cover different layers of
how the interface will be defined. For example, the following layers are
included:


physical, which covers things like the connectors, the pin assignments,
and the electrical specifications. This document is called SCSI Parallel
Interface (SPI).

protocol, which covers all the physical layer activity and how it is
organized into bus phases and packets.

architecture, which provides a description of how the command

requests interact with protocols. This can include things like how the
requests are organized, how the requests are queued, and how each
protocol responds to a request.

primary command, which contains the list of commands that have to

be supported by all SCSI devices.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

The World of SCSI

11

device-specific commands, which would include things like all the

commands that drive CD-ROMs or WORM drives.

Now, when the standards folks started working on this, they recognized
how quickly things were changing, so they layered the specifications to allow
substitution in different parts of the specifications as the technology evolves.
One example would be the standards for the SCSI Fibre Channel interface disk
drive. In this case, the physical and protocol layers would have to be replaced
with new documents, but the other three layers could remain the same.
So, since the newest features are going to show up in SCSI-3, and since
SCSI-3 will be generally higher-performing, you can expect that a SCSI3 device will exhibit better performance than its SCSI-2 brethren. One of
the first things people realized with SCSI-3 was that the number of
peripherals changed. Now, you could have a maximum of 16 devices.
Since there was the possibility of having 16 devices on the chain, the
length of the cable had to increase also. SCSI-3 also saw the added support for a serial interface and for a fiber optic interface. Data transfer
rates depended on the way the hardware was implemented, but the data
rates could actually climb to hundreds of megabytes per second.
Now it is time to get into some of the ways the SCSI-3 standards are
broken up.

SCSI Fast-20 (Ultra)

SCSI Fast-20 is the specification that provides a 20MT/second transfer rate.
That means that the data rate is twice as fast as the SCSI Fast rate. With SCSI
Fast-20, on a Wide bus, the data transfer rate would be 40 Mbytes/second.
This is also referred to as Ultra SCSI.

Dont you wish they would come up with just one name for this stuff and
stick with it?

SCSI Fast-40 (Ultra 2)

SCSI Fast-40 is a set of specifications to define the timing for a future revision of
SCSI-3. This will achieve rates of 40 MT/second, which result in data transfer

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

12

Chapter 1

Disk Subsystems

rates of 80 Mbytes/second. This is referred to as Ultra 2 SCSI. And this is where

things really start to get interesting.
As you may have noticed, each iteration of SCSI had the data transfer
rate doubling. In each case, the speed increase was handled by the bus clock
speed doubling, causing the maximum signaling frequency to double. As
you will see later in the chapter, this was all done using something called
Single Ended signaling. Because the way the information was moved over
the wires did not change, there was backward compatibility. But that compatibility came with a price.
All SCSI data transfers are carried out over cables that are made up of
metallic wires. As signals cross any form of metallic wire, they degrade, and
with this degradation comes a weaker signaland a weak signal is a signal
that is subject to distortion. In order to counteract the weak signal, something
in the SCSI specifications had to change, and usually that was the cable length.
With Ultra SCSI, the cable length had shrunk to just 1.5 meters, which meant
that it would be really difficult to connect 15 devices. Since halving the cable
length again was unacceptable, when Ultra 2 SCSI came out, the signaling
method changed from Single Ended to Low Voltage Differential (LVD).
The switch to LVD allowed the cable length to climb to 12 meters. That
was the good news. The bad news was that moving to LVD required new
dual-mode bus terminators and longer cables. Basically, things stopped
being backwardly compatible. If you happened to attach an Ultra device to
an Ultra 2 cable, the entire subsystem stopped working, completely. People
learned in a hurry to keep Ultra 2 subsystems totally Ultra 2.

Ultra 3 SCSI (Ultra160)

With Ultra 3 SCSI things can get really interesting because the specifications
define as many as 63 variations of features that can be present. A specific set of
these features has been defined by the industry, and that is what is referred to as
Ultra160. The data transfer rate, not surprisingly given the name, is up to 160
Mbytes/second on a Wide bus only.

Ultra320
Ultra320 SCSI is the one that is not off the drawing board yet, but it is going
to feature data transfer rates up to 320 Mbytes/second. Ultra320 was first
defined in SPI-4.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

The World of SCSI

13

Single Ended Devices

Single Ended Devices is really kind of a misnomer, because Single Ended
really refers to the way a SCSI cable is driven by the SCSI devices. Normal
SCSI, or SCSI-1, can also be called Single Ended SCSI. It works like this. Any
time a signal needs to be sent across the bus, there has to be a wire to carry
the signal. With Single Ended SCSI drives, one signal line is grounded and the
other is compared to ground. This affects susceptibility to noise limits, which
impacts things like the maximum allowable cable lengths. So, for example,
SCSI-1 can use cabling up to 6 meters (19.7 feet) long. A Fast SCSI implementation with transfer speeds of up to 10 MT/second can have cabling that
is up to 3 meters (9.8 feet) long. Are you beginning to notice a trend? Think
of it this way: The faster the speed of data transfer, the shorter the cable. For
Ultra SCSI, which will transfer up to 20 MT/second, the cabling can be up
to 1.5 meters (4.9 feet) long if there are more than five active IDs. If there are
less than five active IDs, then the cable can be up to 3 meters (9.84 feet) in
length.

If the cable is not Single Ended, it is Differential SCSI. Differential SCSI comes in
Low Voltage Differential or High Voltage Differential, and these devices are not
compatible on the same bus segment without an electronic device such as a SCSI
converter to convert between Single Ended and Differential. With rare exception,
no software (driver) modifications are necessary for conversion between Single
Ended and Differential. There are several variations of terminators developed for
use with Single Ended SCSI and Differential SCSI.

So, what this means to you the server administrator is confusion. See, the
cable that is used for Single Ended SCSI and the cable that is used for Differential SCSI look the same, even though they are electrically different. To

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

14

Chapter 1

Disk Subsystems

make matters worse, both Single Ended and Differential can use each of the
cable types listed in Table 1.1.
TABLE 1.1

SCSI Cable Type

Cable Type

Characteristics

Type A cable

This is the original SCSI cable. It

contains 50 wires and can be used
by itself for Narrow 8-bit SCSI or in
a combination with a B cable for
Wide 16-bit SCSI.

Type B cable

Added in SCSI-2 to provide a path

for Wide SCSI. It has 68 connections,
and in the early days, it was used
with a type A cable. Since having
two cables proved to be a pain, this
was replaced by the new P cable
defined for SCSI-3.

Type P cable

Instead of containing 50 wires,

type P cable contains 68. It is the
new standard for Wide SCSI
implementations of all speeds.

Firewire cable

A 6-wire cable designed for use with

serial SCSI devices connected by
Firewire. This cannot be used with
any other implementations of SCSI.

Figure 1.4 shows the different types of cable ends for different types of
SCSI devices. About the only way to tell the difference between Single Ended
Devices and Differential Devices is with the judicious use of a volt/ohm
meter.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

The World of SCSI

FIGURE 1.4

15

SCSI cable connectors

Host, or Host Adapter

Connector

Most SCSI Slow (5 Mbytes/sec) computers

and Host adapters use the Centronics-type
50-pin connector. Also, some 8-bit Fast
computers and Host adapters use a
50-pin connector.
Old Sun and DG computers use this 50-pin
connector.
Many 8-bit SCSI Fast (up to 10 Mbytes/sec)
computers and Host adapters use this 50-pin
high-density connector.
Apple/Mac and some older Sun 8-bit
workstations use this 25-pin connector.
All Fast/Wide (16-bit) SCSI-3 computers and
Host adapters, plus old DEC Single Ended
SCSI, use these 68-pin high-density connectors.

Low Voltage Differential (LVD)

With Single Ended SCSI, there is one wire per device. With Differential SCSI,
for any signal that is going to be sent across the bus, there are two wires to
carry it. The first wire carries the same type of signal that the Single Ended
SCSI carries, but the second wire in the pair carries its logical inversion. The
receiver takes the difference of the pair to determine the data. Since it is taking
the difference, you can now figure out the name Differential. This method
makes it less susceptible to noise and allows for a greater cable length.
Low Voltage Differential (LVD) will be finalized in the SCSI-3 specifications, but it will use less power than the current High Voltage Differential
(HVD). LVD is also less expensive than the current technology and will allow
for higher speeds when implemented with Ultra-2 SCSI. The data transfer rate
can be increased to 160 Mbytes/second with a cable length of 12 meters.

High Voltage Differential (HVD)

When you think of HVD, think older technology. HVD supports throughput
of 40 Mbytes/second at cable lengths of 25 meters. These cards and drives
are used in less than 5% of all implementations. They used to be used in
Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

16

Chapter 1

Disk Subsystems

implementations where long cable runs were necessary, especially in noisy

areas. Now LVD can take the place of HVD and save money. HVD is not
compatible with other forms of SCSI devices.
To summarize, take a look at Table 1.2. This covers the types of SCSI.
TABLE 1.2

Different Types of SCSI

SCSI Type

Speed

Connector

SCSI-1
(AKA 8-bit or Narrow)

5 Mbytes/second

Either a 50-pin Centronics, which was standard on things like

scanners or tape
drives, or a DB-25, similar to what is found on
the Iomega Zip Drive

SCSI-2, Fast SCSI

(8-bit Narrow)

10 Mbytes/second

50-pin high-density,
used for things like
Iomega JAZ drives or
writable CD-ROMs.

Ultra SCSI
(8-bit Narrow)

20 Mbytes/second

50-pin high-density,
used for things like
Iomega JAZ drives or
writable CD-ROMs.

Wide SCSI
(16-bit Wide)

20 Mbytes/second

68-pin high-density,
used with hard disk
drives

Wide Ultra SCSI

(16-bit Wide)

40 Mbytes/second

68-pin high density,

used with hard disk
drives

Ultra 2 SCSI
(16-bit Wide)

80 Mbytes/second

68-pin high-density,
used with disk drives

Ultra160 SCSI
(16-bit Wide)

160 Mbytes/second

68-pin high density,

used with disk drives

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

SCSI Termination

17

So, now that you know all about the different types of SCSI, it is time to
ask yourself why it is important. First of all, it is extensible. If you have
worked around networking for any length of time at all, you know that there
is no such thing as too much disk space. If you run out of disk space, it is nice
to know that you can add on another drive, or group of drives, without much
hassle. Cost may be another thing, but without much hassle. There wont be
much hassle as long as you understand termination.

SCSI Termination

long, long time ago, when I first started playing with hardware, the
term SCSI was sometimes enough to bring fear and trepidation into the
hearts of the best of technicians, all because of a couple of small terminating
resistors.
Earlier we mentioned that a SCSI chain has to be terminated at both ends.
That sounds really easy and simple. Sometimes, when you have a combination
of several internal devices connected to several external devices, it is not the
easiest of jobs to locate the end of a chain. In addition, a small resistor that was
plugged into the device terminated some devices, jumpers or DIP switches terminated some devices, and sometimes it was a combination of the two. Some
devices had terminating resistors, which were large and silver and difficult to
lose. So, you always had to remember the basics of SCSI troubleshooting.
Problems are usually caused by termination. When in doubt, break down the
chain and add one device at a time until you find the device that is causing the
problem, or until you get the entire chain working. It could lead to a trying
day. Things have gotten better: Some devices now are self-terminatingthey
just sense if they are at the end of the chain and terminate themselves.
SCSI termination is just the electrical circuitry, which is installed at the end
of a cable that is designed to match impedances for the purpose of preventing
the reflection of electrical signals when they reach the end of the cable. In SCSI,
this is done with a device called a terminator.

When working with any SCSI bus segment, remember there should be
two terminators and only two terminators. Not one, not three, but two
terminators. Also, the terminators must be installed at the very ends of
the SCSI cable, not at devices in the middle of the bus.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

18

Chapter 1

Disk Subsystems

When you talk about SCSI termination, there are four basic types: Passive,
Active, Force Perfect Termination (FPT), and LVD (including LVD/MSE).
Lets explore them one by one:
Passive The simplest form of termination is referred to as Passive. The
terminator consists of a 220-ohm resistor that goes from the TERMPWR
to the signal line and another 330-ohm resistor that goes from the signal
line to ground. This form of termination does not cost much, but there are
disadvantages. For example, if there is a fluctuation in the TERMPWR
voltage, it will show up on the signal lines of the bus. That may be enough
to cause data errors. If your system is using SCSI-2, it is recommended
that you use Active terminators whenever possible for Single Ended SCSI.
Passive terminators are always used with differential (HVD) SCSI.
Active If the termination is not Passive, it must be taking an Active role.
Active termination is referred to as Alternative 2 in SCSI-2. Active termination was developed because of the problems with Passive termination.
To solve those problems, Active terminators have a voltage regulator.
This regulator serves to reduce the fluctuation effect down to practically
nothing. Active termination uses only a 110-ohm resistor, which is
installed from the regulator to the signal line. This provides a much closer
match to the normal impedance of a SCSI cable. This closer match means
a more stable signal, which creates less signal reflection and thus fewer
data errors.
Force Perfect Termination (FPT)
Although FPT is not recognized in
any of the SCSI specifications, it is a Single Ended termination method
that uses diode switching and biasing to make up for any impedance mismatches that exist between the SCSI cabling and the peripheral device,
whatever it may be. Since FPT is not part of the specifications, it should
not come as a surprise that there are several types of FPT and these different types may not be totally compatible. Also, by and large you can
assume that FPT only works and plays well with FPT.
Low Voltage Differential (LVD)
The terminator for LVD uses a form
of Active termination. This termination enhances the faster speeds and
lower power consumption than HVD. It works with Ultra 2 and Ultra 3
SCSI.
LVD/MSE Finally, there is what is referred to as LVD/MSE. This is
LVD that makes use of multimode transceivers. In the case of LVD/MSE,

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

SCSI IDs and LUNs

19

it checks the voltage level appearing on the DIFFSENSE pin of the cable.
By sensing the voltage level, the terminator knows to automatically
configure itself for LVD or for Single Ended. Most new SCSI designs
include these multimode transceivers.
So, now lets see how to put it into action. Take a look at Figure 1.5.
FIGURE 1.5

Cabling external SCSI devices

As you can see, the Host adapter and the last device in the chain are terminated. But what about those things called SCSI IDs?

SCSI IDs and LUNs

have mentioned several times that devices can be linked together in a

SCSI chain. That brings up the question, how does the SCSI controller keep
track of all these devices? Each SCSI device on a chain needs to have a unique
SCSI address, called the SCSI ID. These SCSI IDs are assigned in a variety of
ways, but usually by either a set of jumpers on internal devices or by a rotary
switch for external devices. On an 8-bit SCSI bus, the SCSI ID can be
between 0 and 7. If you are using a 16-bit SCSI bus, the IDs can be between
0 and 15, and for a 32-bit bus, the IDs can be between 0 and 31.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

20

Chapter 1

Disk Subsystems

Remember, SCSI IDs must be unique on the chain. You cannot have two
device 3s on the chain.

How do you choose which address to assign to which device? Lets look at an
example. To keep this simple, lets use an old 8-bit bus because then we dont
have so many numbers to work with. Suppose that we have our controller, three
hard drives, and a CD-ROM. If we are using a regular PC SCSI, we have an ID
range of from 0 to 7. Remember, we are geeks, and all geeks start counting at 0.
In this case, the controller would be set to ID 7, because the higher the number,
the higher the priority. As far as all the rest of the devices, it really doesnt matter
as long as the IDs are unique. For simplicity, we would probably address the
hard drives as 0, 1, and 2 and make the CD-ROM device 3. In this chain, there
would be a terminator on the controller and a terminator on the CD-ROM.

Usually, set the slowest device with the highest number, which will give it the
highest priority. Also, start your numbering at 0. When you boot your system,
the SCSI controller will attempt to contact each device in the chain, starting at
the lowest number. If you have numbered everything from 6 down, you are
going to spend a lot of time waiting for the controller to decide that devices 0
and 1 are not on the chain!

So, you know how to identify the device on the SCSI chain by giving it an
address. What if the device performs different functions, and there needs to
be a way to make that happen? That is where the Logical Unit Number
(LUN) comes into play. The LUN is a value that is used to identify a logical
unit of a SCSI device. According to the SCSI-2 specifications, there can be up
to eight logical units for each SCSI device address. These logical units are
numbered from 0 to 7. To give an example of how this might be used, think
of a tape drive that has a tape changer. In that case, the entire assembly may
have a SCSI ID of 0. The tape drive may have a LUN of 0 and the changer
may have a LUN of 1. Therefore, the actual SCSI address of the tape drive
would be ID 0, LUN 0.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

SCSI IDs and LUNs

21

Bus Length
How long can the SCSI chain grow? As I mentioned earlier, it depends on the
level of SCSI you are using. Check out Table 1.3. It should give you a good
idea of the numbers to keep in mind when working with SCSI.
TABLE 1.3

SCSI Bus Summary

Max
Bus
Length
in
Meters
(LVD)

Max
Bus
Length
in
Meters
(HVD)

Maximum
Number of
Devices
Supported

Bus Speed
MB/Sec
Maximum

Bus
Width
in Bits

Max
Bus
Length
in
Meters
(SE)

Narrow
SCSI-1

N/A

25

Narrow
Fast

10

N/A

25

Fast
Wide

20

16

N/A

25

16

Narrow
Ultra

20

1.5

N/A

25

Wide
Ultra

40

16

N/A

N/A

25

16

Wide
Ultra

40

16

N/A

N/A

Narrow
Ultra 2

40

N/A

12

25

Wide
Ultra 2

80

16

N/A

12

25

16

Ultra160

160

16

N/A

12

N/A

16

Ultra320

320

16

N/A

12

N/A

16

SCSI
Type

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

22

Chapter 1

Disk Subsystems

Servers and SCSI

So, why is all this stuff important? First of all, SCSI is exceptionally versatile.
As a matter of fact, in Minnesota I know of a factory where at one time the
entire manufacturing process was handled from one PC, linked to all the
machines through SCSI interfaces. Now I doubt seriously that you want your
server to handle an entire manufacturing plant, but you probably will want to
take advantage of the way you can add specialty devices like tape backup units
or CD-ROM towers to your network.

Real World Scenario

If you ever get involved in installing SCSI devices, you will find that things can
get confusing quickly. Assigning each drive its own address and making sure
that there are only two terminators on any SCSI chain sounds like such an
easy thing to do, but sometimes, when you are up to your elbows in server
with drives lying all around and jumpers sitting here and there, it can get to
be somewhat out of control.
This is where the KISS rule comes into play. The KISS rule says, Keep It
Simple Stupid. When it is applied to SCSI, it means to take it one step at
a time. If things are not working the way they should, just put one device
into the chain at a time, and get the chain working properly each step
along the way. You will be up and running in no time. The problem with
this little tip is that it requires the technician to have patience, and that is
not usually one of our strongest traits.
When you are dealing with SCSI, especially new implementations, if you
are having problems, think termination and addressing. Those are the most
common problems.

RAID

nother way that SCSI plays an important part for your server is in the
way it can be used to provide redundancy and increased performance. Much
of that is done with a technology called Redundant Array of Independent
Disks (RAID).

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

RAID

23

In every book I have ever written for Sybex, I have mentioned the gee-whiz factor of computers and of networking. The gee-whiz factor works like this: I understand how it works, I know why it works, I even know how to make it work, but
the fact that it works the way it does still amazes me. Now, I admit, I am easily
amused. But when I think about routing, I am truly amazed. When I think about
the elegant simplicity of Domain Name Service (DNS), I am amazed. But I am
really in awe of RAID technology. The upper levels of RAID are seriously impressive. So, this is what we are going to talk about in this section.

Definition of Terms
I know, I know, there is a whole glossary in the back of the book dedicated
to defining terms. I also know that there are terms that I am going to be
throwing out over the next several pages that we should come to some sort
of a common understanding about. Not that you would fail to take the time
to look in the back of the book to see what they meanthat would never
happen.
So, when we start talking about things like RAID, we start using terms
like high availability or fault tolerance. High availability means just what it
sounds like, making sure that the resources your server provides are available a high percentage of the time. Fault tolerance means that if something
breaks, there is something else there to pick up for the broken part, and
things go on as if nothing happened.
Another term we should look at is the phrase single point of failure. A
friend of mine says that you can tell the skill level of a network administrator
by her level of paranoia. The really paranoid ones are the ones who have
been around the block and understand that the question is not if something
is going to go wrong, but when. They also know that no matter how bad they
think it can get, it can get worse. In any computer system, there are components that can go bad. The reliability factor is getting much better, but stuff
still does happen. You are trying to increase your odds, so that when things
do go bad, you are covered. You know that certain components in a system
have a higher chance of failure than others. For example, it is much more
likely that a printer is out of paper or is offline than that the mainboard in
the printer has gone bad. So, by looking at where our single point of failure
is, we are hedging our bets.
Here is a brief example. If I provide a level of disk drive protection called
mirroring, it means that two disks are hooked to a single controller. Everything that is written to one of the disks is written to the other disk. If one of
the disks goes bad, the other disk is there to take over for it, and we have
fault tolerance and higher availability. We have, in effect, moved our single
point of failure from the hard disk back to the disk controller. You can move
Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

24

Chapter 1

Disk Subsystems

the single point of failure back farther than that, but that would be stealing
my thunder for the section on duplexing.
Next, there is the subject of data striping. With data striping, instead of
taking bits of information and storing it on disk, you are taking bits of information and storing it across several disks. In this way, write heads on several
disks are being utilized and performance increases dramatically. Unfortunately, there is no fault tolerance, so if any one of the disks in the stripe set
goes bad, the entire set is dead in the water. This is not necessarily a good
thing, so there is something called striping with parity.
Finally, we get to the subject of parity. What follows is a highly simplistic
explanation of parity, but it should give you an idea of how it works. First,
take a look at Figure 1.6.
FIGURE 1.6

Disk striped with parity

Disk 1:
30GB

Disk 2:
30GB

Disk 3:
30GB

Disk 4:
30GB

Disk 5:
Parity storage

Assume that each one of the first four disks is 30GB each. The fifth 30GB
drive is not used to store datait is just used to store the mathematical sum
of the information striped across the first four drives.
So, now we are going to save a file called RESUME.DOC to the striped set of
drives with parity. In this case, lets assume that the first block of data can be represented in binary as 1010. That means that a 1 would be written to Drive 1, a
0 written to Drive 2, a 1 written to Drive 3, and a 0 written to Drive 4. Finally,
the sum of 1+0+1+0 or 10 (remember, we are dealing in binary here, and in
binary 2 is represented as 10) would be written to Drive 5. Parity is defined simply as the quality of sameness or being equivalent. With RAID we are using parity not only to check to make sure things are the same, but we are also using it
to rebuild things that may have been damaged. Take a look at Figure 1.7 to see
what I mean.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

RAID

FIGURE 1.7

25

Disks striped with parity and data on the drives

10

Finally, it is the middle of thunderstorm season, and you take a massive

lightning strike. The only thing that gets damaged is Disk 4, and it is pretty
much finished. See what I mean in Figure 1.8.
FIGURE 1.8

Disk striping with parity and a dead drive

10

Because you have instituted parity, life can go on without anyone being
the wiser. If someone wants to access the file RESUME.DOC, the system recalls
the file and loads the 1+0+1. It knows there is supposed to be something in
place of Disk 4, but since it cant find it, it reads the parity sum of 10 and
knows that all that is missing is another 0. The system can continue functioning until you get another drive installed. Hopefully, there will not be any
more thunderstorms!
So now that we have terms defined, we can talk about RAID.

The Levels of RAID

We are going to talk about seven of the more commonly defined levels of
RAID. Some of these may already be in place on your server, or none of these
may be in your servers, and you may have different implementations in the
same server.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

26

Chapter 1

Disk Subsystems

RAID 0
This is disk striping without parity. Of all the RAID technologies, RAID 0 is
the fastest because the write heads are constantly being used with duplicate
data being written, or without any parity being figured. With this system your
server will have multiple disks and the information is striped across the disks
in blocks without parity. There is no fault tolerance in a RAID 0 system.

RAID 1
This level is commonly referred to a disk mirroring, or disk duplexing. In either
case, there are two hard disks involved and anything that is written to one of the
hard disks is written to the other. In the case of disk mirroring, there is just one
disk controller, so the controller is the single point of failure. Disk duplexing
adds a second controller to the second disk, moving the single point of failure
away from the disk subsystem to the mainboard. In RAID 1, if either disk fails,
the other disk takes over. There is no parity or error checking information
stored. If both drives fail, new drives must be installed and data restored from
backup.
The disadvantage of RAID 1 is cost per megabyte. If you have two drives
that have a published capacity of 30GB each, and they are mirrored or
duplexed, the total amount of usable disk space is 30GB, not the 60GB you
purchased. If you are using different-sized drives, the mirror will reflect the
storage capacity of the smallest drive.

RAID 5
In this case, data and parity information is striped at block level across all of
the drives in the chain. Again, RAID 5 takes advantage of the faster disk
reads and writes. The parity information for data on one disk is stored with
data on another disk, so if any one of the disks fails, it can be replaced and
the data can be rebuilt from the parity data stored on the other drives. Again,
it requires a minimum of three drives, but usually five or more disks are used.
The disadvantage here is that the controllers are becoming expensive.
Reasons to Use RAID 5
When you start talking about RAID 5 and higher levels, the hardware
controller can become something of an issue. This can cause the price
point of the implementation to climb. Obviously, you are going to use it
on mission critical servers like these:


Enterprise critical file and application servers

Database servers

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

RAID

Web, e-mail, and news servers

Intranet servers

27

RAID 0+1
Now we start getting into some of the hybrid approaches. If you look at the name,
and you understand what RAID 0 and what RAID 1 do, you have a pretty good
idea of what RAID 0+1 is. You know that RAID 0 is disk striping without parity.
You know that RAID 1 is mirroring or duplexing of disks. So, RAID 0+1 is where
an entire stripe set without parity is actually mirrored or duplexed. There will be
a giant performance improvement on disk reads, there will be some performance
hits on disk writes. Data will survive the loss of multiple disks, but the monetary
cost can be high.

RAID Setup, Uses, and Recovery

We are about to start bouncing around some of the objectives, so be prepared.
We are going to look at the differences between Hardware RAID and Software
RAID, and what happens when a drive fails. That subject is going to introduce
some more terms to remember, things like hot swap, hot plug, fail over, and
hot spare.
One of the things that any really good server administrator takes into consideration is what happens when things go wrong. Now, most of the levels
of RAID that we have talked about have built-in levels of redundancy somewhere. There is only one level (RAID 0) where the loss of a single drive will
result in data loss. In all the other types, losing a single drive, or even losing
multiple drives, will not result in down time. So how is RAID implemented
and what happens in each case when a drive fails?

Software RAID
RAID 0 and RAID 1 are usually defined at the software level. In this case, it is
the server operating system that determines the RAID level and the level of protection. In Windows NT/2000 it can be called Drive Striping, Drive Striping
with Parity, or Mirroring. In NetWare it may be called Mirroring, but the result
is same. Somewhere there is a tool or utility that will allow you to either stripe
a drive and provide parity or mirror the drives.
The advantage of using Software RAID is low cost. There are no special
controllers to buy. The operating system will recognize the drives and provide
the level of protection that you define.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

28

Chapter 1

Disk Subsystems

Hardware RAID
In some of the more complex implementations of RAID, a special controller or
special disks need to be linked together. When you start mentioning the word
special, the dollar signs usually start to light up. It will be up to the controller
to define the level and type of RAID.
Because you are dealing with hardware rather than software, your
performance will increase.

Hot, Hot, Hot

ou have hot swaps, hot plugs, and hot spares. What in the world is
the difference and how do they work? It is all a matter of degree!

Hot Spare
A drive is considered a hot spare if you happen to have an extra drive sitting on
the shelf that matches the type and configuration of the drives in your server. For
example, if you have a SCSI-2 Ultra, 9GB Seagate on the shelf waiting in case of
emergency, that would be considered a hot spare. When the hot spare gets put
into play, it could be a hot plug or a hot swappable drive.
Here is an example of how a hot spare would be used. Say your server has
RAID 1level mirroring defined. The first drive in the mirrored pair has
failed, for no other reason than drives go bad, and the system has failed over
to the second drive in the mirror. In this case, the system keeps on working
like nothing has happened. You notice the fail over (the fact the first drive
failed and the second took over) and make plans to replace the failed drive
with a hot spare when the server can be taken out of service with a minimum
amount of interruption to the normal workday. When you can down the
server, you shut it off, replace the failed drive with the hot spare, and bring
the server back online. Once the server is online, you can then use the appropriate tool to reestablish the mirror, and the new drive will be mirrored to
match the old.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Hot, Hot, Hot

29

Hot Plug
With a hot plug drive, the server does not have to be brought down or taken
out of service to install a new drive. In the case of a hot pluggable drive, you
are not replacing a current drive that has failedyou are adding disk space to
the mix. In the case of a hot pluggable drive, you open a cabinet, plug the drive
into the backplane of the cabinet, and the operating system should recognize
the drive is there. Depending on the operating system, you will have to create
a partition and a volume to make the drive available.

Hot Swap
This is one of those gee whiz things we talked about earlier in the chapter.
I remember the first time I had to hot swap a drive in a RAID array with parity.
When I asked a senior tech how to do it, he smiled and said, Open, pull, push,
watch, and be amazed. I got to the client site with the drive in hand, and went
to the server room. There was a large disk array of seven drives in a cabinet
with a glass door, all sorts of flashing lights next to six of the drives, and a
series of steady red lights next to the drive that had died. It didnt take a rocket
scientist to figure out which drive had failed. So I opened the glass door and
saw the two small rocker arms holding the bad drive in place. I moved those
out of the way, grabbed the handle on the front of the bad drive, pulled, and
the drive came out in my hand. I took the new drive, pushed it gently into the
slot until I felt it lock, and then put the rocker arms back in place. Once that
was done, the lights next to new drive started going crazy, while the drive was
automatically rebuilt from the other drives in the set.
It was seriously cool! No one on the network had any clue that a drive had
ever failed! No data was lost, no time was lost, and the server was never
unavailable.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

30

Chapter 1

Disk Subsystems

We probably should have gone over this earlier, but we didnt! As you read
through many of these chapters, keep in mind that there are several laws of
network computing that come into play. Some of these are documented,
some are only figments of my imagination, but they are important to remember just the same. Here are some of my favorites: Williamss Law: You can tell
the skill level of a network administrator by his level of paranoia. The really
good ones are really paranoid. Murphys Law: Anything that can go wrong,
will go wrong, at the worst possible moment. Govanuss Law: The chance of
completing any network upgrade successfully is inversely proportional to the
visibility of the project and the proximity of your annual review. If you are
about to undertake a project that will affect everyone on your network, and it
is the night before your annual review, please be sure to be carrying a copy of
your resume on a disk in your pocket. You may not make it back to your desk.
Finally, my favorite, and this one has proven true worldwide: End users lie.
Network administrators are the best end users.

Real World Scenario

The concept of hot plug, hot swap and hot spares is certainly something
you should take into account if your budget will allow. We are all striving for
the longest possible uptime, though we all know that stuff will happen. Any
time you can cover yourself and your key components without having to
down the server to make a repair, your users come out ahead.
When you look at the key components that are most likely to suffer a failure,
most can fall into the hot plug, hot swap category, and all it takes to make
that happen is money! Not a problem, right? As a consultant, I am routinely
called in to explain why the increased costs for a specific product are worth
the money. I tend to use an approach I learned a long time ago when I was
a salesperson. I try to talk the managers language. I use terms like cost of
operation. Managers know what it costs to operate a department or a company down to the penny. I can take those figures and extrapolate them to
help someone make a decision. For example, if it costs the company $1,000
an hour to operate your department, and if we agree that it will take a minimum of four hours to change out a key component, the cost to the company in downtime is $4,000. If there is an added cost of $500 to make this
device hot swappable so there will be no loss of service, we can show where
it is a sensible decision to spend the extra money.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Fault Tolerance

31

Fault Tolerance

Fault tolerance is the act of protecting your computing gear, whether that
gear be infrastructure-oriented as in switches and routers or computer-oriented
as in servers and disk farms. In either case the fundamental question you ask
yourself is this: How can I protect the equipment so that a fault of some kind
doesnt interrupt service? Impairment of service might be tolerable: interruption
is not.
We talk about uptime of devices and services in terms of 9s. We assume
that you want to keep your gear up 99% of the timethats a given. But as
we add 9s to the other side of the decimal point, the time that a device is
allowed to be down, including maintenance windows, becomes increasingly
smaller. Five nines uptime equates to an allowance of about 4 hours
downtime per year, including maintenance windows. Your goal with fault
tolerance methodologies is to increase the number of 9s that are on the right
side of the decimal point. Five nines is optimal, but not realistic in most situationsfour nines is a better goal. Well talk about how to realize these
goals in this chapter section.

Configuring RAID
One thing you can do is use RAID to help augment your system uptime.
Either RAID 1 (disk mirroring) or RAID 5 (disk striping with parity) will be
beneficial to you in terms of bringing fault tolerance to your servers.
Some network operating system software allows you to set up RAID configurations without having to purchase special hardware RAID array controller cards (theyre expensive). I have worked with software RAID and dont
think it works very well. I much prefer hardware solutions. For starters the
card has its own processor and memory and can really go a long way in offloading the central CPU from having so much work to do. With software
RAID the CPU handles everything. Also, the software solution seems to not be
as reliable as the hardware-based solutionthough it may have been my fault
configuring the software rather than how the software behaved. Whatever the
reason, Im saying that when you consider RAID implementations, you should
pay the extra $2K or so and get the RAID controller card with the system.
Another important point about hardware RAID is that there is always
some data in the cards memory. If you had an ungraceful shutdown on the
server while there was some data in that card, it would be lost. Thus its
important to purchase your RAID array controller cards with a battery

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

32

Chapter 1

Disk Subsystems

backup so that in the event an instantaneous down happens to the server, the
data will be safe for a time until you can bring the server back up. Keep in
mind the data is being held there by a battery, so you dont have days or anything like that, but you do have some cushion you can work with.
Youll usually opt for either a mirroring or a striping-with-parity scenario
for a given set of drives. You can have both kinds in your system without
encountering any difficulty at all. As a general rule of thumb, you usually
want your OS to be on a mirrored set of disks while your data will live on a
RAID 5 volume. Some NOS software wont work on a RAID 5 volume at all.
Youll typically configure the RAID volumes through either a BIOS
interface at the cards boot time or through a configuration CD that
comes with the server. HP, for example, includes a wizard-like interface
that you can use to configure the entire box, including the RAID array.
Watch the BIOS messages at boot time and youll be given the key
sequence to enter so that you can access the cards BIOS.
Theres also the concept of a RAID 10 where you configure two separate
drive cages with RAID 5 arrays and then mirror the arrays. Youve got
double fault tolerance because if the first array has two drive failures, you
can break the mirror and work on the second drive array until you get the
first one fixed.
Realize that just because the systems on RAID doesnt necessarily mean
itll never have to be taken down. RAID helps safeguard systems so that they
can keep working until users go home and you have a chance to down the
computer and make repairs after hours. You want to avoid downing servers
during working hours.

Adding Drives to the Array

Adding drives to the array is very easy to do. First you physically add the
drive, then you access the arrays BIOS (or use the configuration CD) and
add the drive to the volume that you want it to be a member of.
There is one minor thing to think about and that is that the arrays must
use the same amount of drive space off of each drive in the system. So if you
have an array that consists of three 4.3GB hard drives and two 5.7GB drives,
the array will set up a RAID 5 volume that uses five 4.3GB pieces of space,
one from each disk. Thatll leave 1.4GB for you to do something else with.
The RAID card will show that space hanging out there and available elsewhere. You cannot include disparate lengths of data in a RAID 1 or RAID
5 volume. You can opt to configure the leftover space anyway you like
whether its one unprotected volume of space, a mirror, or however you

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Fault Tolerance

33

want to configure it. Keep this in mind as you add disks to the system. You
cannot add a hard drive that is smaller than the current array is expecting
and be able to configure a volume. You must provide as large a disk or larger
in order to facilitate the addition. If you have some left over, its up to you
to configure the extra space as you see fit.
Lastly, remember the n-1 rule with RAID 5. You take the number of disks
youre going to dedicate to the RAID 5 array and subtract 1 from that number
to account for the space needed for the parity stripe. Thus if you have six
17GB hard drives youre putting in an array, youll really only wind up with
5 * 17GBs worth of data because you sacrifice one disks worth of space for the
parity stripe. Actually the stripe is usually sent across all the disks so youre really
not dedicating one disk to parity stripe, though there are RAID implementations
that will allow you to do such a thing. With RAID 5 then, more disks means that
you attain more actual disk storage space and dont sacrifice as much space to
parity striping. More is more in the case of RAID 5.

Replacing Existing Drives

Youre usually alerted to hard drive problems in RAID arrays through events
being posted to log files in the NOS system logs. Typically its incumbent
upon you to set up the software needed to perform this monitoring of drives
and posting to logs, so be sure you read your system documentation.
If a drive goes bad in a RAID 1 or RAID 5 array, generally you live with the
problem until you can obtain a like replacement and then shut the machine
down after hours to fix the problem. You then perform a graceful shutdown,
replace the drive, and then go into the array controller cards configuration
utilities (through the cards BIOS or the configuration CD) to rebuild the stripe
or rebuild the mirror, depending on which failed. Its a very straightforward
process.

Performing an Upgrade Checklist

As always, when youre performing replacements or adding disks to an
array, its wise to take a little bit of time to plan out your activities and jot
them down in a checklist. Be cognizant of the need for additional drivers or
software. Be sure to read FAQs and readme docs before proceeding. Its not
a bad idea to have a lab environment where you can test some of these kinds
of activities before you ever have to do them in real life, thus giving yourself
some practice opportunities.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

34

Chapter 1

Disk Subsystems

Planning for Redundancy

The key word to keep in mind when youre considering highly available,
highly fault tolerant implementations is redundancy. Thats how the phone
company does itthats how you have to do it if youre looking for five nines
uptime.
When you order servers and you have an eye toward redundancy, there
are many key places to look:
Hard drives You can consider multiple hard drives in a single RAID
environment, multiple RAID environments, and hot spares.
Power supplies Consider purchasing multiple power supplies with servers. If a single power supply goes, you have a redundant unit that can keep
the server going.
Cooling fans Server cooling is of paramount importance. Consider
multiple cooling fans in servers that you order.
Network Interface Cards Dual-homing or multiple-homing a server by
adding a second or even more NICs provides a way for you to add redundancy to the server. If a single NIC fails, users can still get in because
theres a second one in the box. Dual-homing can also be used to provide
a server path for multiple IP segments.
Multiple processors Adding more than one processor to a computer
allows for symmetric multiprocessing (SMP)aware applications to be able to
utilize all of the processors in the system. It also provides redundancy in the
event any one processor fails.
Uninterruptible Power Supply Hooking your servers up to a UPS,
whether its a stand-alone UPS designed to keep just one server up or a big
room-size unit that can handle hundreds of servers. If the power goes out
in your building, the UPS will hold the servers until you can perform a
graceful shutdown.
Generator For long-term outages many companies have a fossil fuel
powered generator that can start up and continue to supply power to the
computer room until power is restored.
Clustered servers Having two computers connected to a single data
source and running cluster software allows for the second computer to pick
up and continue operations in case of the outright failure of one computer.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Fault Tolerance

35

Planning for fault tolerance is all about redundancy. You should consider
applying redundancy in any of your mission-critical servers. A little bit of
money spent now can save countless hours of downtime later on.

Using the Technique of Hot Swap

As a busy server administrator youll doubtless run into times when you have
a hard drive fail in a RAID array and you have to replace the drive. This is
called swapping the drive (go figure). There are three kinds of swapping:
Cold swap A cold swap means that you have to actually power down the
computer in order to perform the swap. Software-based RAID 5 would put
this kind of requirement on you. Youd go into the server logs, see that the
RAID software was alerting you that a drive had gone offline, and then
youd have to schedule maintenance time to down the server and replace the
drive. Once the new drive was installed and the server brought up, the
RAID software may have to be manually put into rebuild mode or it may
automatically rebuild the stripe for you.
Warm swap Warm swap is in between a hot swap and a cold swap. In
a typical warm swap you dont have to power down the computer, but
you may have to instruct the RAID controller to stop allowing I/O transactions to the array while you change out the drive. The controller does
this by simply placing any requests for I/O on hold until the drive has
been replaced and is ready for operation. The advantage is that you
avoid delays that are incurred by drive spin-up, controller boot-up, and
negotiation with the CPU. Youll typically see warm swap scenarios in
PCI-based RAID systems.
Hot swap Hot swap allows you to replace a faulty hard drive without
having to power down the computer or introduce a pause to I/O requests
(with the exception of disallowing attempts to write to the bad component, that is). You can replace the drive and bring it online without very
little intervention with the other working components and no disruption
to service. Hot swap technology ensures high availability in RAID arrays
and, best of all, allows admins to service the unit during working hours
without having to schedule an after hours downtime period for the computer. Be advised that its still advisable to perform this kind of work after
hours because you just dont know what kind of effect youll have on the
server. But its good to know that in a Web server that has to be up
24x7x365, for example, a hot swap would allow you to replace a hard
drive without users seeing a blip on the radar screen.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

36

Chapter 1

Disk Subsystems

Theres one other technique thats often useda hot spare. In a hot spare
situation, you keep a spare drive in the computers drive bay. You configure
the RAID array controller to treat the drive as a spare. When data is written
to the stripe, the drive is included as a backup drive. If one of the main drives
in the array fails, you can utilize the hot spare to act as a fallback. Hot spares
are handy because you simply have to go into the RAID configuration utility
and tell it to begin using the hot spare. The downside is that you burn a hard
drive you wouldnt ordinarily have to use.
All of the above systems provide high-availability scenarios in the case of
a single drive failure. Keep in mind that two or more drives failing means the
end of one array.

Creating a Disaster Recovery Plan

As an admin, you should, sooner rather than later, ask yourself this question: If a disaster of incredible significance hit this buildingi.e., a flood,
earthquake, explosion, or other cataclysmic eventhow would I restore
the operations of the business back to normal? Furthermore, how soon
could I have them restored?
This is the question that Disaster Recovery (DR) asks and presents the
problem that you must solve. At first the DR questions are intimidating.
But DR plans are like eating an elephantone bite at a time. To facilitate
a good DR plan you must examine several elements of your network
design:


How reliable are my backups?

Am I storing my backup tapes offsite?

Have I performed a test restoration from one of my backup tapes?

Do I have adequate replacement computers and network gear to

rebuild the current facility?

Do I have an offsite DR facility that I can roll the business to in the

event of a disaster?

How many hours or days behind can my business be in the event of a

catastrophic event and subsequent DR restoration of service?

One thing that should be obvious from looking at the above list is that you
cannot make DR decisions alone. Clearly youll need to solicit the advice and
interaction of others in order to facilitate a robust DR plan.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Fault Tolerance

37

Remember the basic concepts of DR: fault tolerance, the ability to gracefully recover from a fault, and redundancy. If, for example, your business
runs entirely off of Web activity, then your Web servers are of paramount
importance to you. So much so that you cannot afford for them to go down.
In such a case, a DR plan might include the following components:


Redundant ISP to act as a backup to the primary ISP

A multiplicity of Web servers all running some sort of Network

Load Balancing (NLB) software so that not any one server is used
inordinately more than the others

Redundant WAN links to your ISPs

Redundant DNS servers for name-server backup

Security software that monitors for hacker, denial-of-service attacks,

and other unscrupulous events

Primary, secondary, and tertiary backup administrators for the Web

servers

Clustered computers that can allow for the failure of any one server

Hopefully you get the idea. DR means that you provide an offsite place
where a redundant copy of your operation can live in case the first instance of
your operation somehow gets annihilated. Redundancy means that you build
fault tolerance into the feature set so that you avoid annoying little failures that
have the capability of driving the entire enterprise to its knees. You put this all
down in writing in a DR plan and then you periodically test the plan to make
sure it works for todays operations.

Identifying Hot and Cold Sites

In DR terms, a hot site is your primary operational site and a cold site is the
backup site that will become hot in the event something catastrophic happens
to the first site.
DR is expensive. Thats because you have to supply redundant everything
from servers to peripherals to methodologies to sites. In highly mission-critical
computing scenarios where the servers just cannot go down (think phone company, think stock exchange, think national security), you must think about
ways to provide a secondary location that can very quickly become available
in the event the first site becomes suddenly (and most likely permanently)
unavailable.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

38

Chapter 1

Disk Subsystems

Mainframe computing centers have long relied on redundant cold sites

that maintain exact copies of the primary computing environment. And
they rigorously test their DR methodology using the cold site as the fallback site for operations. There are companies that function as DR cold site
facilities for various companies around the world. In the server arena, site
DR is something that people are just now beginning to talk about.
Suppose you have an e-mail server that cannot go down. You might think
about introducing a cluster server scenario where one of the servers in the cluster lives in your primary building, but the other one lives offsite in a different
building, perhaps not even a building you own. There is some sort of a highspeed copper or fiber-optic link between the servers so that the secondary one
is aware of the primary one at all times and can pick up operations in case the
first one fails. This is a very high-level idea that requires much more intensely
technical understanding of cluster and database technologies, but you see the
point. You have a cold site thats all prepped and ready to begin operation
should the hot site fail.
Keep in mind that cold sites are designed to work temporarily. Typically
theyre not designed with permanence in mind. The thinking is that youll
eventually fix the hot site and restore it back to normal operations.

Summary

o, what we have done here is give you some protection against Murphys Law, and a way to prove that you match Williamss definition of a
really good network administrator. There is more to be done on this front,
but that about takes care of the disk subsystem. First we are going to do some
review, and then in Chapter 2, IDE Devices, we will be looking at clustering, Fibre Channels, CPUs, and multiprocessing.
We also talked about fault tolerance and all of its nuances. There is one
basic notion that comes into play when we think about fault tolerance:
redundancy. The acronym RAID, for example, stands for Redundant
Array of Inexpensive (or Independent) Drives. You use redundancy to
build in high-availability. A hot spare drive is one that sits in the drive cage
and can be put into play by tweaking the RAID utility. Hot swap capability
means you can change out a hard drive without any interruption to users.
Warm swap means you can change out the drive in an array, but you have
to disrupt I/O requests long enough to get the drive replaced. You avoid the

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Summary

39

drive spin-up, controller boot, and other associated nuisances of a cold

boot, but users do suffer a lapse in servicea bad no-no in fault-tolerance
terms. A DR plan details how youll put your companys computing back
in business after a disaster of some kind occurspristine as it was the day
before the disaster happens. To do this youll denote a hot and a cold site.

Exam Essentials
Know the difference between logical drive and physical drive A physical drive can contain multiple logical drives but a logical drive will usually reside on one physical drive.
Know the different levels of raid, and what makes each level unique
RAID 0 is disk striping without parity, RAID 1 is disk mirroring or
duplexing, RAID 5 has data and parity information striped at the block
level across the drives, and RAID 0+1 is where a disk array that has been
striped without parity is also duplexed or mirrored.
Know which levels of scsi can interoperate without an adapter and which
levels will require an adapter SCSI, SCSI-2, and Ultra SCSI all use a
50-pin connector that is interchangeable. Wide SCSI, Wide Ultra SCSI,
Ultra 2, and Ultra 160 use a 68-pin connector.
Know the appropriate length of the various scsi cables SCSI is 6
meters, Fast SCSI is 3 meters, and Ultra SCSI is 1.5 meters with more than
five devices. If there are less than five devices, then the cable can also be
3 meters in length.
Be comfortable with the differences between hot plug, hot spare, and hot
swap. A hot spare is a device that is waiting to be put into the machine.
The other two choices are very close in meaning: A hot pluggable device
is one that can be installed while the computer or server is turned on, and
a hot swappable device is one where the device can be removed and
replaced and the server will experience no loss of service. For example, a
single network card can be hot pluggable. Drives in a RAID 5 array can
be hot swappable. If one of the drives fails, it can be removed and
replaced, and the data can be rebuilt on the fly without any loss of service.
Know how to configure drives. Be able to add or change drives in an
array and configure accordingly.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

40

Chapter 1

Disk Subsystems

Key Terms
Before you take the exam, be certain you are familiar with the following terms:
Active
American National Standards Institute (ANSI)
asynchronous mode
data striping
disk duplexing
Domain Name Service (DNS)
Fast SCSI
Fast-Wide SCSI
fault tolerance
Force Perfect Termination (FPT)
high availability
High Voltage Differential (HVD)
hot spare
jumpers
Logical Drive
Low Voltage Differential (LVD)
LVD/MSE
Mega Transfer (MT)
mirroring
parity
Passive
Physical Drive
RAID 0
RAID 0+1
RAID 1
RAID 5

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Summary

Redundant Array of Independent Disks (RAID)

SCSI Fast-20
SCSI Fast-40
SCSI Narrow
SCSI-2
Single Ended signaling
Small Computer System Interface (SCSI)
termination
Ultra 2 SCSI
Ultra SCSI

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

41

42

Chapter 1

Disk Subsystems

Review Questions
1. What is the width of the data transfer bus of SCSI-1?
A. A nibble
B. 4 bytes
C. 8 bytes
D. 8 bits
E. 16 bits
2. SCSI-1 is also referred to as which of the following?
A. Narrow
B. Slow
C. Fast
D. Wide
E. Ultra
F. Ultra 2
G. Narrow, Fast, and Wide
3. Choose one type of connector used in SCSI-1.
A. 9-pin serial
B. 15-pin serial
C. 25-pin Centronics
D. 50-pin Centronics
4. RAID stands for which of the following:
A. Redoubtful Array of Inexpensive Diskettes
B. Redundant Array of Inexpensive Disks
C. Redundant Array of Independent Disks
D. A SWAT team action

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

43

5. Which is not an ANSI standard?

A. SCSI-1
B. SCSI-2
C. SCSI-3
D. All of the above
6. Horace is the network administrator for a large dot-com. One day the

hard drive in one of the Web servers fails. The server is running hardware
SCSI-based RAID. What kind of drive changeout can Horace most likely
perform?
A. Cold swap
B. Warm swap
C. Hot swap
D. Hot spare
7. Ultra and Ultra 2 are examples of which of the following:
A. RAID 10
B. SCSI Bus Width
C. Physical Drives
D. SCSI Bus Speed
8. With LVD SCSI, how many wires will be dedicated to carrying the signal

across the bus?

A. 1
B. 3
C. 2
D. 4
E. 8
F. 16

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

44

Chapter 1

Disk Subsystems

9. What is the maximum Bus Speed of SCSI-2?

A. 5.0 to 20.0 Mbytes/second for a 16-bit data bus
B. 10.0 to 15 MT/second for an 8-bit bus
C. 2.0 to 4.5 Mbytes/second for a 16-bit data bus
D. 40 to 60 MT/second for a normal 8-bit bus
10. You are examining the possibility of putting a SCSI-2 controller in

your server. You still need to keep a device that works with normal
SCSI. Is it possible to run the SCSI device from the SCSI-2 controller?
A. No, SCSI-2 is not backwardly compatible to SCSI.
B. No, SCSI is Single Ended, and all SCSI-2 is HVD.
C. Yes, SCSI-2 is backwardly compatible with SCSI.
D. Check proper termination.
11. What is the maximum number of devices that can be part of a SCSI-3 bus?
A. 16 devices
B. 8 devices
C. 7 devices
D. 14 devices
E. 15 devices
12. What is another name for SCSI Ultra?
A. SCSI Wide
B. SCSI Fast and Wide
C. SCSI 20
D. SCSI 40
E. SCSI Fast 20

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

45

13. How does Single Ended SCSI work?

A. For any signal that is going to be sent across the bus, there are two

wires to carry it. Both wires can carry the signal.

B. For any signal that is going to be sent across the bus, there are two

wires to carry it. One wire will carry the signal and the other wire
will carry a defining voltage.
C. For any signal that is going to be sent across the bus, there are two

wires to carry it. One wire will carry the signal and the other will
be ground.
D. For any signal that is going to be sent across the bus, there are two

wires to carry it. Both wires alternate with the signal.

14. Choose one way in which RAID 3 differs from RAID 1.
A. With RAID 3, the data is always written with two types of parity.
B. With RAID 3, the data is striped in bits.
C. With RAID 3, the data is striped in blocks.
D. With RAID 3, the data is striped in bytes.
15. How does RAID 5 write data to the disk?
A. In bits
B. In nibbles
C. In bytes
D. In binary
E. In blocks
16. Can a hot swappable drive also be a hot spare?
A. Yes.
B. No.
C. It depends on the level of RAID.
D. It depends on the speed of the SCSI devices.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

46

Chapter 1

Disk Subsystems

17. You have a SCSI controller in your server and now you wish to add

some new SCSI-2 drives to this system. Can you do so?

A. Yes, but you will have to buy a new controller.
B. Yes, SCSI and SCSI-2 devices are compatible.
C. No, SCSI-2 is not compatible with the earlier SCSI specification.
18. Which of the following solutions combines data striping across drives

with mirroring?
A. RAID 1+5
B. Hybrid RAID 5+
C. High Performance RAID
D. RAID 0+1
19. Which of the following SCSI standards can use a 68-pin connector and

therefore will be interchangeable?

A. SCSI
B. SCSI-2
C. Wide SCSI
D. Wide Ultra SCSI
E. Ultra 2
F. Ultra 160
G. Ultra SCSI

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

47

20. Your boss has asked you to implement Hardware Level RAID because

he understands that it is more reliable. He wants your opinion. What

will you tell him? Select two.
A. It is not necessarily more reliable but it does provide better

performance.
B. Software Level RAID is less expensive but provides better performance

because the process is controlled by the network operating system.

C. Hardware RAID will cost more because he will have to purchase

a special controller and disks.

D. Software RAID will end up costing more because a special version

of the operating system has to be purchased.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

48

Chapter 1

Disk Subsystems

Answers to Review Questions

1. D. The data transfer bus in SCSI-1 is 8 bits.
2. A, B. SCSI-1 is also referred to as SCSI Narrow or SCSI Slow.
3. D. SCSI-1 uses a 50-pin Centronics cable.
4. C. If you answered B, you are showing your age. When RAID tech-

nology was first introduced, it was described as a Redundant Array

of Inexpensive Disks. Now, however, it has been changed to the
Redundant Array of Independent Disks.
5. C. At the writing of this book, SCSI-3 was still a proposed standard.
6. C. Because the system is SCSI-based, it likely supports the use of hot

swap. If Horace has an extra drive sitting around (one that has the slide
rails used for his computers drive cage), all he has to do is pop the old
drive out, put the new one in, and the RAID controller should automatically take over. Some older controllers require you to manually begin
the array rebuild.
7. D. Ultra and Ultra 2 are examples of Bus Speed.
8. C. With Low Voltage Differential, any signal that is going to be sent

across the bus will have two wires to carry it.

9. A. The maximum Bus Speed of SCSI-2 is 5.0 to 20.0 Mbytes/second

for a 16-bit data bus.

10. C. Yes, it is possible, but it is not recommended.
11. A. A SCSI-3 bus can have 16 devices.
12. E. SCSI Fast 20 is another name SCSI Ultra.
13. C. For any signal that is going to be sent across the bus, there are two

wires to carry it. One wire will carry the signal the other wire will be
attached to ground.
14. D. With RAID 3, the data is striped in bytes.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Answers to Review Questions

49

15. E. RAID 5 data is striped at block level across all of the drives in the

chain.
16. A. Whether a drive is hot swappable has nothing to do with its status

as a hot spare.
17. B. SCSI-2 was backward compatible with SCSI, but for maximum

benefit, it was suggested that you stick with one technology or the
other, preferably using a SCSI-2 controller with SCSI-2 devices.
SCSI, SCSI-2, and Ultra SCSI all use a 50-pin connector that is
interchangeable.
18. D. RAID 0+1 is a hybrid approach where an entire stripe set without

parity is actually mirrored or duplexed.

19. C, D, E, F. Wide SCSI, Wide Ultra SCSI, Ultra 2, and Ultra 160 use a

68-pin connector.
20. A, C. Hardware RAID costs more because of the special controller

and disks that need to be purchased but it provides significantly better

performance than Software RAID. In addition it will support Hot
Swap of drives that fail.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Chapter

IDE Devices
SERVER+ EXAM OBJECTIVES COVERED IN
THIS CHAPTER:

3.3 Add hard drives.


Verify that drives are the appropriate type.

Confirm termination and cabling.

For ATA/IDE drives, confirm cabling, master/slave, and

potential cross-brand compatibility.

Upgrade mass storage.

Add drives to array.

Replace existing drives.

Integrate into storage solution and make it available to the

operating system.

Perform upgrade checklist, including: locate and obtain

latest test drivers, OS updates, software, etc.; review FAQs,
instructions, facts and issues; test and pilot; schedule
downtime; implement using ESD best practices; confirm that
the upgrade has been recognized; review and baseline;
document the upgrade.

3.6 Upgrade adapters (e.g., NICs, SCSI cards, RAID, etc.).



Perform upgrade checklist, including: locate and obtain

latest test drivers, OS updates, software, etc.; review FAQs,
instructions, facts and issues; test and pilot; schedule
downtime; implement using ESD best practices; confirm that
the upgrade has been recognized; review and baseline;
document the upgrade.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

t wasnt that long ago that SCSI drives were selling for about
$1,000 a gigabyte and memory was selling for $100 a megabyte. Maybe it
wasnt that long ago by the calendar, but in computer terms, it was eons
ago. In the late 1980s, Integrated Drive Electronics (IDE) drives were introduced as a lower-price-point alternative to SCSI drives, or some of the
other high-priced, low-performance alternatives. Since the 80s IDE drives
have come a long way, to the point where they are being shipped in an estimated 90% of all systems sold. Lets take an in-depth look at the history
of IDE drives, how they have overcome some of the barrier limitations
imposed by the technology, and where IDE technology is today. We can
then take a look at the types of cabling and connectors required to install
IDE devices, and some of the differences between IDE and SCSI.

For complete coverage of objective 3.3, please also see Chapter 1. For complete
coverage of objective 3.6, please also see Chapters 1 and 10.

IDE Disk Drives and Subsystems

s you remember from taking the A+ exam, disk subsystems are made
up of the hard disk, the cabling, and the disk controller. Last chapter, in our
discussion of SCSI, you saw how the controller had to be matched to the type
of SCSI technology, and how the controller played an active part in moving
data and instructions. Disk controllers can be integrated into the mainboard,
or they can be on a board that plugs directly into the mainboard. Sometimes
these are called controllers, but you may also see the terms paddle cards or
even paddleboards.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

IDE Disk Drives and Subsystems

53

The reason they are referred to as paddleboards is simple. The boards do

not do any controlling. All controlling functions are handled by the electronics
on the drive, hence the name Integrated Drive Electronics (IDE). The original
version of IDE drives helped increase the amount of storage space the average
user could purchase without filing for bankruptcy.

Like many of the computer industry acronyms, IDE has picked up several
definitions. Depending on the book you read, it may be Integrated Device
Electronics, or it may be referred to as Integrated Drive Electronics. This falls
under the tomato (toe-may-toe)/tomato (toe-mah-toe) argument; it really
doesnt matter much where the acronym came from as long as you know
what it is referring to and the darn things work.

Since the drive was controlled by electronics on the drive, the drive
manufacturers could encourage enhancements, because there were no
pesky controller compatibility issues to contend with. Each manufacturer
was free to include some new techniques that would increase capacity,
speed, and the average time that the drive could operate without failure,
called the Mean Time Between Failure (MTBF). Some of these advances
included error checking, or the ability to automatically move contents
from blocks that were failing to blocks that were specifically set aside for
the purpose, or generating higher disk rotation speeds to ensure faster
data access, and even giving the user the opportunity to re-map the drive
geometry if desired. Lets take a look at the history of IDE to track where
it has been until today.

IDE and ATA

It seems like the terms IDE and AT Attachment (ATA) are joined at the hip.
If you see IDE, you are more than likely to see ATA, and if you see ATA, it
is some form of IDE device. ATA is really the formal title of the standards for
IDE drives and how they operate. IDE is kind of like Kleenexit is more of
a trade name that refers to a 40-pin interface and the integrated controller
design. The official name should really be ATA devices.

IDE hard drives, but those are by no means the only devices that take advantage of IDE technology. There are also things like IDE tape devices and IDE CDROMs.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

54

Chapter 2

IDE Devices

As we mentioned above, IDE was originally designed so the disk controller was integrated into the drive itself. This meant that the drive no longer
had to rely on a stand-alone controller board for instructions as all of the
other types of drives did. This integration brought the cost down. It also
made the drives firmware implementations easier to manage for the manufacturer. This meant you had a device that didnt cost very much, and was
exceptionally easy to install. People loved it and the boom of the disk drive
industry was on.

ATA History
When ATA was introduced in the late 80s, it was a hard drive only type of
technology. At the time the ATA standard was approved, applications and
operating systems came on diskettes and only the real computer aficionado
had a CD-ROM device. Most CD-ROMs at the time were SCSI based and
expensive. Since there werent many, if any, things being distributed on CDs,
the CD-ROM was not a necessity for most folks.
As applications and operating systems grew, diskette distribution
became unwieldy, not to mention expensive. There had to be a better
way, and that better way was to distribute software on CDs. After all,
CDs were very inexpensive and could hold almost 700MB of data. CDs
were relatively impervious to end users. For an end user to do something
to damage a CD, they had to work pretty hard.
Now it became imperative that a reliable, low cost method be made available to distribute CD-ROM drives to the masses. The designers of the ATA
specifications suddenly needed to come up with a way to attach things like the
CD and the various tape drives or other storage devices on the existing disk
subsystem. Using the same ATA controller card to manage two devices would
be infinitely more viable than having to put yet another controller card in an
already crowded computer bus. So, the designers came up with something
called the ATA Packet Interface (ATAPI). ATAPI is a fancy name for an extension of the ATA interface. The extension is designed to allow several other
types of devices to plug into an everyday, ordinary old ATA 40-pin cable.
There are some differences in the way ATA supports hard drives and the way
it supports other devices. The hard drives receive support through the system
BIOS. It is up to the BIOS to define the geometry of the drive. These other devices
required a special device driver to support them. So, for example, if you had
installed an early version of the SuperWhizBang 8 X CD-ROM, you would originally need a driver from SuperWhizBang so the system would recognize the fact
that the drive was there. Back in the old DOS days, this required editing the
AUTOEXEC.BAT and CONFIG.SYS files to make sure everything worked just the

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

IDE Disk Drives and Subsystems

55

way it was supposed to. Depending on the operating system you are using, there
may still need to be some manual configuration of devices.
The standards continued to mature, and CD-ROM manufacturers started
working together to provide support for ATAPI. As ATAPI drives became
more standardized, operating systems, and in many cases the BIOS, were
able to recognize the CD-ROM. If the O/S or the BIOS could recognize the
drive, it could immediately load the driver, and if the BIOS can recognize the
CD-ROM, the CD can even be used as a bootable device. This eventually led
to some new advances that we take for granted today, with things like CDROMs that will autorun programs to start installations.
Back to the good-ol-days. When CD-ROMs became viable, it brought up
another shortcoming of the early ATA standard: That was the number of
devices you could have in an ATA chain. With the early drives, you could have
a maximum of two drives connected to a paddleboard and there could only be
one paddleboard in the computer. As you will see, the later implementations
of the standard increased the number of ATA channels in any machine to two,
so you can now have up to four ATA devices in a system. We will discuss how
to configure those four devices a little later in the chapter, in the section on
master/slave/cable select.

IDE, the Next Generation

So, we move on in the history of computing to the early 90s. The ATA architecture wasnt keeping up with the advances in hard drive technology. The
hard drive industry came up with a new standard, and for the lack of a better
name just called it ATA 2. ATA 2 was a great improvement over ATA because
it defined faster data transfer modes using Programmed Input/Output (PIO)
and Direct Memory Access (DMA). PIO is a way of moving data from the
storage device so that all the information has to be sent through the processor.
Different levels (called modes) define PIO, and each mode has its own data
rate. This is shown in Table 2.1.
TABLE 2.1

PIO Levels and Data Flow Rates

Mode

Rate

3.3 MBytes/sec

5.2 MBytes/sec

8.3 MBytes/sec

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

56

Chapter 2

IDE Devices

TABLE 2.1

PIO Levels and Data Flow Rates (continued)

Mode

Rate

11.1 MBytes/sec

16.6 MBytes/sec

As the need for speed increased, the PIO standard couldnt keep pace. That
was when DMA came into being. Instead of the device sending information
through the processor, now the information was written directly to memory.
Because the information is written directly to memory, the Central Processing
Unit (CPU) doesnt have to do anything with it, so the overall performance of
the computer is increased. DMA and Ultra DMA can increase processing
speeds to 100 MBytes/second, but we are getting ahead of ourselves.
Back to ATA 2. In addition to the different methods of handling data,
there were many other under-the-hood kinds of things that the average user
probably wouldnt be aware of. These included things like some powerful
drive commands, like the Identify Drive command. This command was a
godsend to technicians everywhere. Prior to the standardization of the
Identify Drive command, the technician who installed the drive had to
know some exact information on the way the drive was configured. That
usually wasnt a problem if it was the original installation of the drive and
you had all the documentation right there, but if the drive were ever
moved, or pulled out of one machine to be used in another, the configuration information tended to get lost. (Not that something like that would
ever happen to menope, never happen, because I always write the drive
specifications on the outside of the drive with a permanent marker. And if
you believe that, let me know, I have a great bridge to sell you just outside
of McCausland, Iowa.) Then you had to search for the documentation in
your exceptional filing system or call the manufacturer.
Anyway, that problem went away with the updated drives and updated
BIOS. If the drive was an ATA 2 device, when the drive was installed in a
computer, you simply had to install the drive and turn the computer on.
The BIOS would go out and discover the drive automatically. The drive
tells the BIOS how it is built and then the BIOS makes sure the rest of the
computer knows how to address the drive and how much viable space there
is. It is a wonderful thing. It is really one of the first instances of Plug and
Play, only this installation happened well before the operating system even
started to load.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

IDE Disk Drives and Subsystems

57

Another advance was the way the drives handled the data transfer. Instead
of moving the information bit by bit, or even byte by byte, ATA 2 began to
allow block data transfers, called block transfer mode. Think of it this way:
Imagine you have just gotten back from the grocery store after buying one
months worth of groceries for a family of four. Further imagine that all you
could carry into the house was one item at a time. That would take you a really
long time to get everything into the house. That is the way it was before block
transfer mode came into play. Now, with block transfer mode, compare how
much more efficient it is to carry the groceries in one or two sacks at a time.
It may still take you a while to move all the stuff into the house, but not as long
as the other way. Block transfer mode just moved more information in a single
operation.
These block transfers were made possible by a new way of defining and
addressing the sectors on the hard drive. This was done using a process called
Logical Block Addressing (LBA). LBA had an additional benefit, because it
managed to overcome the early IDE size limit of 528MB.
ATA 2 maintained its backward compatibility with ATA drives. It used
the same 40-pin physical connector used by ATA, and an ATA 2 drive could
be used in conjunction with an ATA drive.
There are some other ways that ATA 2 may be described. For example,
you will hear terms like Enhanced IDE (EIDE) or Fast-ATA. Each of these
is not a standard, but just a different implementation of the ATA 2 standard.
EIDE, which started out as a particular manufacturers implementation, has
become so popular that EIDE has become more or less a generic term.
ATA 2 also introduced the capability of having two channels of two
devices per paddleboard. This meant that the total number of IDE devices
that were possible in a system had climbed to four. The channels were
referred to as the primary channel and the secondary channel.

ATA 3
The next standard is ATA 3. ATA 3 does not do anything for the faster transfer modes, but it does provide for password-based security and better power
management. It also has a technology called Self-Monitoring Analysis and
Report Technology (SMART). SMART will tell you when a drive is going
bad before it exhibits any symptoms that you may be aware of.
If you sometimes wonder why your computer takes a long time to respond
after you have let it sit for a while, ATA 2 may be part of the reason. You see,
it also added some sophisticated power management features that would put
the drive to sleep after it hadnt had anything to do for a while. ATA 3 is also
backwardly compatible with ATA 2, ATAPI, and ATA devices. You may also

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

58

Chapter 2

IDE Devices

see the term EIDE applied to ATA 3 devices, since there has been no significant
improvement in data transfer.

ATA 33 or Ultra DMA 33

Ever faster means ever better, and one step forward with ATA devices was
with the ATA 33 standard. This is where the ATA standard began moving
away from the PIO standard and began taking advantage of DMA.
The ATA 33 specification provided high-performance bus mastering that
gave 33 MBytes/second DMA data transfer. Now, bus mastering sounds like
a real geeky term that must have some deep inner meaning. When you get
down to it, there really isnt much mystery. Bus mastering is a technology
that the drive or controller can use to direct the way data traffic is routed
through the input/output path. In the case of ATA 33, the bus master sent the
information directly to memory, rather than to the processor as had been the
case in previous installations.
Since there were new ways of doing things, there had to be some way
of designating the new technology. Lets face it, ATA 33 is not really
all that sexy, so this implementation can also be called Ultra DMA 33
or just UDMA 33. If you install one of these devices, you will need an
ATA 33 drive, controller, and BIOS support to receive the full benefit
of the technology; however, it is fully backwardly compatible. Like all
the previous implementations of IDE, it can use a 40-pin IDE-type cable
unless one of the following is true:


You are using a low-quality, damaged, or weakened cable.

The system has excessive signal noise caused by multiple drives, a dual
power supply, or even an integrated Cathode Ray Tube (CRT).

The system has been put in overclocking mode, or has been set beyond
the manufacturers specifications.

ATA 66
Well, if ATA 33 moved data at 33 MBytes/second DMA, you will never guess
what rate ATA 66 moves data. You got it. It uses even faster high-performance
bus mastering for a 66 MBytes/seconds DMA data transfer rate. This can also
be called Ultra DMA-66 or just UDMA-66.
If you are going to install an ATA 66 drive, you will need the appropriate
drive, controller, and BIOS. Again, it is fully backwardly compatible with the
previous ATA standards, but the cabling has changed. The change was necessary because the transfer rates became so high that there needed to be more
protection against things like crosstalk and electromagnetic interference

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

IDE Disk Drives and Subsystems

59

(EMI). To handle this problem, 40 additional conductors were added as

ground lines. The ground lines act as shields between the lines that carry the
real signals back and forth. The bottom line is you will still need a 40-pin
cable, but now it will have 80 conductors.
The computers operating system must also support these rates of data
transfer.
ATA 66 Issues
If you are installing this type of disk subsystem in your server and run into
problems, here are some things to check:


Make sure you have the right cable. You can tell you are using a
40-pin/80-conductor cable because it will have a black connector
on one end and a blue connector on the other end, with a gray
connector in the middle. The blue connector goes to the motherboard, the gray connector is for the slave device, and the black
connector is for the master drive. In addition, the cable has something you probably wont be able to see: Pin 34 should be notched
or cut. The reason will become plain in the next bullet.

The motherboard or mainboard controller must be capable of supporting the ATA 66 standard. A compatible controller has a detect circuit that can recognize the fact that line 34 is not present on the cable.
If the detect circuit is missing, the motherboard may be able to detect
the presence of an ATA 66 cable, but may try to configure the device
for a higher transfer rate.

Some controllers may not be able to handle the ATA 66 on both the
primary and secondary channels. If you are having problems installing
the device on the secondary controller channel, you may want to move
it to the primary channel and see if that solves the problem.

Make sure you have the right controller card driver. Make sure the
BIOS is upgraded, and any patches that need to be applied to the
motherboard have been taken care of.

Be sure you are using a DMA-capable operating system and that the
DMA mode has been activated.

Make sure the drive has been configured to run at ATA 66 transfer
rates. Some drives ship with the higher transfer rate disabled by
default; enabling the higher transfer rate is done with either a jumper
switch or with a software setting.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

60

Chapter 2

IDE Devices

ATA 100
The most recent advance in the world of ATA/IDE is the release of the ATA
100 interface. As you can tell from the name, the ATA 100 specifications
allow for the transfer of data at a rate of 100 MBytes/second. This is the
transfer from the host-to-drive bus. The new interface does maintain some of
its history, using the same 40-pin, 80-conductor cable as the ATA 66. This
means that like all the other devices we have talked about so far, the ATA
100 cable can be used with other, slower drives. These can include things like
hard disks, removable media disks, CD-ROM drives, CD-R/RW drives,
ATA tape drives, and DVD-ROM drives.
There are other advances made with ATA 100. One is something that has
been around the computer world for a while: Cyclic Redundancy Check
(CRC). The CRC is a very-high-level method of checking to make sure the
transferred data actually made it through the transfer process without
becoming corrupted. It is just a data reliability check.
It works like this. When the device that is transferring the data gets ready to
send it, it attaches an extra set of bits to every frame of data. These extra bits are
called the Frame Check Sequence (FCS), which acts as a type of verification that
is attached to each frame. When the frame is received, the receiver does the math
and checks to make sure the answer is what it expects. If it does, all is good. If
it doesnt, the frame has been corrupted and it needs to be retransmitted.
Lets look at a really simple example. Remember when you were kids and
had those really cheesy secret decoder rings that came in cereal boxes? That
way you could send messages to your friends, and if the teacher intercepted
them, she couldnt read them out loud. Well, the basis of that was usually
some kind of mathematical formula. We will assume that the sender is going
to multiply everything by 3 and that the receiver knows that. So, we take a
look at a simple four-bit frame:
1101
Now, since we are working with a frame, that is a binary number, not
a decimal number, so 1101 translated from binary to decimal is 13. Since
we agreed we are going to multiply everything by 3, our 13 becomes 39.
Converting that to binary, we have this result:
100111
Now, we are going to make another assumption, and we are going to
assume that our packet is made up of two parts; the first contains the answer,
and the second part contains a sequence of three sets of 10 and then the data.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

IDE Disk Drives and Subsystems

61

When I send this to you it is going to be the answer, 100111, followed by

101010, followed by 1101. So, you will get a string that looks like this:
1001111010101101
When you receive it, you break the transmission down by removing the
101010 and keeping everything else. So you are left with 100111, which
you divide by 3. You come up with 1101, which you compare to the data portion. Since the two parts are equal, life is good and you can go on to the next
transmission. If it is not good, you send a message back to me, telling me to
retransmit the information. Simple, huh? Why is such a simple technique
popular? Well, it has three things going for it. First of all, as you have seen,
there is the way it can detect errors even in extreme circumstances. Secondly,
the whole process does not add very much overhead and, because it is a mature
process, it is relatively easy to implement.
As is the case with some of the other ATA specifications, in order to get the
most out of ATA 100, you should have a controller that meets the ATA 100
specifications. This controller can be either a controller that is integrated into
the motherboard, or it can be external. Your hard disk and other devices
should support the ATA 100 interface and you should have one of the 80-conductor cables that supply grounding.
You may also see the ATA 100 interface referred to as the Parallel ATA,
since all data is being transferred in parallel. This means that the system can
transfer several bits of data at the same time. While this seems efficient, there
are some drawbacks. First of all, IDE is designed to be less expensive than
SCSI, but that 80-conductor cable is expensive to manufacture and it takes up
a lot of space in the PC case. Because the cable (actually any IDE, SCSI, or
floppy drive cable for that matter) is bulky, it can block or inadvertently redirect airflow from the fan, causing the system to overheat. As you know, when
we start talking about Parallel in the world of computers, that means that there
may also be a Serial, and that is what is just around the corner. Several manufacturers are looking at getting much faster transfer speeds for the IDE standards, and this is just not going to happen with the current ATA design.

What Is Next for the ATA Standard?

The latest and greatest for ATA is called the Serial ATA. There is a consortium
of manufacturers working together to bring the new standard to life. These
manufacturers include some of the biggest names in the industry, like APT
Technologies, Dell, Intel, International Business Machines (IBM), Maxtor,
Quantum, and Seagate.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

62

Chapter 2

IDE Devices

This group is shooting for a new interface that will increase throughput to at
least 160 MBytes/second, with later versions reaching 528 MBytes/second. In
order to do this, the cable design is going to have to be radically altered. Instead
of the current 40/80 cable that allows for only four attachments, the new cable
will be much smaller, with only four signal pins and a few more pins for power
and electrical ground. What is this going to do to the current technology?
According to the Frequently Asked Questions (FAQ) at the Serial ATA
Working Groups Web site (www.serialata.org), the new implementation is
going to be designed so that it will drop into a PC and be compatible with the
software, meaning it will run without modification to your current computer
(other than the appropriate controller and devices). Since the cable will be
smaller, they will be easier to route and easier to install.
What about all your old stuff? It is anticipated that there will be a period
where both the old parallel standard and the new serial standard are available.
Now, this could cause a problem. Since both types of devices are going to show
up in the same machine, and since each will have its own interface, the Serial
ATA group expects that there are going to be some adapters to adapt the serial
cable to be able to handle the old 40/80 devices.
Serial ATA is going to support all the normal ATA and ATAPI devices,
including CDs, DVDs, tape devices, high capacity removable devices, and
Zip drives. One of the other goals is to make the devices easier to upgrade,
because the Serial ATA group is planning on eliminating jumper settings for
defining the devices role.

For more information on jumper settings and drive roles, see Master/Slave/
Cable Select and Jumper Settings later in this chapter.

Serial ATA is not planned to be a threat to technologies like USB because

Serial ATA is going to be internal to the computer only. There is not going
to be any type of external interface for either PC storage devices or other
peripherals.

Cabling and Connectors

s we have gone through the discussion of ATA devices, you will

notice that there were basically three types of connectors. The earliest was
the 40-pin connector, which handled things up to ATA 66. With ATA 66,
the throughput was so great that 40 more conductors were needed to lead
into a ground to help prevent crosstalk and EMI. How can you tell them
apart? Take a look at Figure 2.1.
Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Cabling and Connectors

63

Now if you look closely, you can notice several things. The first is the
thickness of the conductor channel. With the 40-pin connector, the channel
is much thicker than it is with the 40/80 connector.
FIGURE 2.1

ATA IDE cables

ATA/33 cable
40 conductors

ATA/66 cable
80 conductors

The next thing I would like you to notice is that there is a dark line down
the right side of each cable. This line indicates the location of Pin 1. That will
become important in just a few seconds. Each of these cables, although you
cannot see it, has two other similar connectors on it. One of those connectors
would attach to the controller and the second connector would attach to
another ATA device.

You may be asking yourself why there is only one extra connector. Remember
that with IDE, unlike SCSI, there can only be two devices in a chain; SCSI can
have seven. Depending on the IDE controller, there can be up to two IDE chains
in any computer, for a total of four devices. Also, SCSI can handle external
devices, while ATA cannot.

Lets talk about installation. First of all, take all the usual precautions.
Turn the computer off, and unplug it. Always work with an antistatic mat
and an antistatic wrist strap. The antistatic mat is made of a conductive
material that is set on the top of your worktable, and then the computer or
other component is set on the mat. When the antistatic wrist strap is fastened
to the mat, the electrostatic charge level of anything placed on the mat will

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

64

Chapter 2

IDE Devices

become equalized with the charge level of the mat, and these will become
equalized with the charge level of your body. After the charges have been
equalized, electrostatic discharge (ESD) sparks will not occur. Now, lets
assume that you are installing a device that has the controller for both channels built right into the motherboard. We will also assume that you have
already mounted the device in the case. The first thing you have to do is
attach the cabling to the motherboard. Remember that colored stripethis
is where it comes into play. Since we are going to be adding another device
to the system (yes, this is another assumption), you locate the connector for
the second IDE channel. It should be marked on the motherboard with something really creative like IDE-2. Then, looking very carefully at the motherboard, you will see a small 1 near the end of one of the connectors. That
shows you where Pin 1 is on the motherboard. Now, Pin 1 on the motherboard has to match Pin 1 on the cable or things just will not work. Once you
have located Pin 1 on the motherboard, carefully line up the holes on the
connector with the pins on the motherboard, keeping the 1s together. Push
down gently until the cable is snug to the motherboard.

Here are a couple of tips. First off all, be careful to make sure that the pins are
all lined up with the holes on the connector before pressing down too hard. If
you should happen to bend or break one of the pins, it will probably ruin your
day. That is especially true if the controller is embedded in the motherboard.
That would mean replacing the motherboard, usually an expensive proposition. Secondly, once the cable is firmly attached to the mainboard, take a permanent marker and mark the channel in big bold numbers, so the next time
you have to add something to the IDE chain, you can immediately know which
channel you are dealing with. The channel information is silkscreened on the
motherboard, but I usually need a flashlight and a magnifying glass to read it.
This way is just simpler.

Once the cable is attached to the motherboard, you can attach the cable
to the device you are going to install. Again, check to find Pin 1. If you cant
find Pin 1, look closely at the male connectors on the drive. There will usually be either a space without a pin, or there will be a notch in the plastic connector sleeve. Check the end of the cable, and you may see one of the
pinholes blocked, and you may also see a notch on the cable. Line those up,
plug the cable in, and seat it firmly. Plug the power from the power supply
into the device and you should be ready to power up the computer. The
ATA-66 cable is keyed. Remember the blue keyed end attaches to the
motherboard. If the standard ATA cable is installed in Reverse Pin 1 to Pin
40, the hard drive LED will stay on continuously.
Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Master/Slave/Cable Select (CSEL)

65

How do you know if it is there? Well, depending on the computer you are
using, watch what happens when the system boots. Some BIOS implementations will show you the devices they find as they go through Power On Self
Test (POST). Otherwise, you may have to access the BIOS to see if the device
has been recognized, or, depending on the operating system, the new device
may be visible through something like Windows Explorer.

Real World Scenario

ATA/IDE installations, especially of a single device on a channel, are really
painless. They usually go easily. If, however, things are not working as
planned, here are a several things to check. First of all, does it have power?
I have done a great job of installing devices only to forget to plug in the
power lead from the system power supply. If it has power, check to make
sure the ATA connector cables are securely fastened to the device and to
the motherboard. You may need to unplug them and re-seat them. Dont
try to tell me they look snug, just unplug and re-seatit saves time arguing. If things still arent right, recheck the Pin 1 to Pin 1 issue. Those are the
three most common (and most embarrassing) issues you can have with
ATA installations.

Master/Slave/Cable Select (CSEL)

ll through this chapter, we have been mentioning the fact that there
can be two devices, and only two devices, in an ATA subsystem. When you
start examining the advantages and disadvantages of IDE versus SCSI, that
is just one of the areas where IDE falls short. The other area is in the way the
two devices are linked together.
As you know by now, IDE stands for Integrated Drive Electronics. All of the
drives intelligence is on board every single drive. That is a great thing if you
have only one drive on the subsystem, but when there are two drives hooked
together and both want to be the brains of the operation, things dont work well.
With IDE devices, you have to relegate one of the drives from being the brains
of the operation to being the go-fer. It is called designating one of the drives to
being the master, and one of the drives to be the slave.
So, there are two ways a single channel of IDE components can be strung
together. Take a look at Figure 2.2.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

66

Chapter 2

IDE Devices

FIGURE 2.2

IDE drive configurations

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Master/Slave/Cable Select (CSEL)

67

As you can see, in the top part of the diagram there is a single drive
attached to the IDE host adapter. In the bottom part of the drawing, there
are two devices taking orders from the same host adapter.
Defining the master and the slave is done with jumpers. Now, there are
three possible settings:


Default: master with no slave present

Master with slave present

Slave with master present

Jumper Settings
Figure 2.3 shows what the business end of an ATA device looks like. If you
look closely at the picture, you will see that there are three sets of pins circled.
FIGURE 2.3

Business end of an IDE/ATA device

Jumpers

You will also notice that there is a small piece of plastic covering two of the
pins. This very small but very powerful tool is called a jumper. You see, each
set of pins represents a channel that the information signal can take from the
controller to the electronics on the drive. The presence or lack thereof of the
jumper completes a circuit that defines the path the electrical impulses will
take. For example, if there were no jumpers present, the information would
follow the path so the drive would be configured as a master, with no slave
device present. Look closely at Figure 2.4 to see the different types of settings.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

68

Chapter 2

IDE Devices

FIGURE 2.4

Master/slave jumper settings

In Figure 2.4, the master slave selection switch is designated as J8. If there
were no jumpers present, the drive would be configured as a master, with no
slave device present. Having a jumper covering Pins 3 and 4 of Switch J8 may
designate the drive as being the master with a slave present. The other drive in
the chain would then have to have a jumper covering Pins 5 and 6, indicating
that it would be the slave device, taking all of its instructions from the master.

If your controller or motherboard supports two ATA drive channels, each

channel can support a master and a slave device. Each channel must have the
master and slave devices properly configured, or that channel will not work
properly.

Now it would be a wonderful thing if I could tell you that each set of pins
for the master/slave relationship was labeled J8, and that in each and every
case, no jumpers indicated master, jumpers across 1 and 2 indicated master
with slave present, and jumpers across 3 and 4 indicated slave with master
present. It would be a wonderful thing, but it would not be the real world.
Now, while it is generally true that the absence of a jumper usually indicates
a master with no slave present, in the real world, things just may not be what
they seem. When you are configuring any ATA devices, be sure to check the
Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Master/Slave/Cable Select (CSEL)

69

appropriate documentation. If you cant find it, check the Web. Be sure to
check the documentation before you start removing jumpers and putting
jumpers back on again. Trust me, it will make your whole life a lot easier.

If you have installed multiple ATA devices, and one of them is recognized by the
system and the other isnt, or if neither of them is recognized by the system, shut
the machine off and start over. Your jumpers are in the wrong place. If you get
things really flummoxed, you may want to go back to the beginningin other
words, install the first device as a master with no slave. Check to make sure it is
recognized. Remove the first device, and install the second device as a master
with no slave; check to make sure that it is recognized. Once that has been done,
you know both your devices are good. Then configure one as the master and the
other as the slave and install them. Check to make sure they are both recognized.
If not, check to make sure the cable is tight. If the cable is tight and one (or both)
of the devices is still not being recognized, and you are absolutely positively certain the jumpers are 100% correct, replace the cable. Make sure you replace the
cable with the right type cable for the most advanced type of ATA device in the
chain. In other words, if you have an ATA 66 device in the chain, you should be
using a 40/80 cable. In this scenario, the potential problem areas are the jumper
settings and the cable. As a last resort, replace the jumpers. Sometimes they lose
the metal sleeve that covers one of the pins, and contact is not made.

Cable Select
Now, if you have been really sharp, you will have noticed that there were three
sets of pins shown in Figure 2.3, and up until now, only two sets of those pins
had a reason to be jumpered. The third set is for cable select (CSEL), which
does just what it says, it lets the controller decide which drive will be the master
and which drive will be the slave. CSEL is one of those features of the ATA
specifications that have been around for a while, but you may have never had
an opportunity to work with it. There were some problems with the original
specifications. Look at Figure 2.5, which shows that CSEL has one drive
added, and it has been assigned the drive letter C:.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

70

Chapter 2

IDE Devices

FIGURE 2.5

CSEL with letter C: assigned to drive

Typical 1-drive cabling using old CSEL

Drive 1
C:

Look at Figure 2.6 to see what happens when you add a second drive.
FIGURE 2.6

CSEL with two drives assigned

Typical 2-drive cabling using old CSEL

Drive 0
C:

Drive 1
D:

You will see that the cable select has automatically assigned the letter D: to
the second drive. Now, if you have completely used all of the space on the first
drive for C: and the second device is a CD-ROM, all is fine. What happens if
you havent? Say you have created a drive and assigned the letter D:. Now you
have chaos.
Some controller manufacturers have done serious work to solve the problem,
but for the most part, especially in a server implementation, you may want to be
really sure and just take matters into your own hands and configure the settings
yourself.

SCSI versus ATA/IDE in a Server

This is one of those discussions that is bound to border on religion and

politics. Some people will say never put an IDE drive in a server. Others will
say they work just fine, and with the high MTBF, massive capacity, and
lower cost, why not? So, always being the diplomat, I will say, it depends.
It depends on the implementation and also on the depth of the pockets of
the company buying the server. SCSI offers superior redundancy as well as
unlimited expansion. ATA/IDE cannot match that, yet, just by the limitations
placed on it by the technology. SCSI is also expensive.
Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

SCSI versus ATA/IDE in a Server

71

Analyze the implementation. If you have something that needs 99.99999%

uptime, look to SCSI and RAID arrays or duplexing at a minimum. If you have
a server that is not mission critical, and cost is a major factor, look to ATA and
a really good backup system.
Dont get caught in the ATA disk-mirroring dilemma. Some folks will say,
Hey, we are using two IDE disks that cost us $150 apiece for 60GB. We
have just used software mirroring, and so we have a set of mirrored disks for
about $300. Its true, you have mirrored disks, until one of them fails. If the
drive fails that is the slave, the system will perform as advertised. If the drive
fails that is configured as the master, your mirror has just lost all intelligence
and the system is down. At this point, mirroring ATA/IDE drives on the same
channel just does not work.

Real World Scenario

As I mentioned in the chapter, installing IDE devices has really become
close to a no-brainer. The most difficult thing you will have to do is make
sure that you have the master and slave setting correct. The cables will
usually have either a notch that fits into an appropriate spot on the drive,
or one of the pins will be blocked out, making it almost impossible to plug
the cable into the drive the wrong way. That leaves connecting the cable
to the motherboard or to the controller.
Each cable will have a colored stripe down one side. This indicates Pin 1.
When you look at the connector on the motherboard or on the controller,
you will see a small 1 silkscreened somewhere on the silicon. Match the
stripe with the 1 and you should be in good shape. Another way to tell Pin
1, if you have really good eyes, is to look at the solder connection of the pin
to the motherboard. You will notice that all the solder connections are
round, except one. That one is square, and that indicates Pin 1. This can be
described as a fallback if you cannot locate the printed information.
If you are installing devices, there are certain indications that things are not
necessarily correct. For example, if you are adding a CD-ROM to an IDE
chain, when you turn the system on, the power light on the CD-ROM should
come on briefly and then go off. If the light stays on, you either have the
cable reversed or you have misconfigured the master/slave settings.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

72

Chapter 2

IDE Devices

If you are installing new devices into a computer, and things are not working
as planned, the first thing to check is the configuration of master and slave.
After that, check to make sure the cables are plugged in properly.
Oh, yeah, one other thing. Dont be like me. I tried for about 15 minutes to
get an IDE CD-ROM to be recognized by the system before I noticed that
while the master/slave was right, and the cable was in the right way, having
a power cord connected should have also been a priority. There are times
we all do really dumb stuff, and I think I hold the record!

Adding Hard Drives

dding hard drives to a computer can be extremely easy or can present

you with numerous questions and difficulties. The funny part is that its your
choice as to whether things will be easy or not. You must understand the
choices you have relative to hard drives and when to pick which selection.
You must also strive to get some hands-on experience with various hard
drive installations so youre very comfortable with any situation youre
thrown into.

Verifying That Drives Are the Appropriate Type

You first must determine whether youre dealing with an Integrated Drive
Electronics (IDEalso called Intelligent Drive Electronics) or a Small Computer
System Interface (SCSI) hard drive.

The term AT Attachment (ATA) is synonymous with IDE. ATA has gone
through several version iterationsmostly due to increased computer bus
speeds. Visit www.webopedia.com and perform a search on the keyword ATA
for more information.

The easiest way to tell the two apart is to simply look at the connector
cables for each. IDE uses a 40-pin connector and SCSI uses anywhere from
Centronics and DB25 (for SCSI I) to a 50-pin connector for SCSI II and a 68pin connector for SCSI III. Theres no mistaking an IDE cable for a SCSI
cable. So, when in doubt, even if the hard drive doesnt have a label or a cable

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Adding Hard Drives

73

you can count the number of pins it will accept and youll know what kind
of drive youre dealing with.

Note that its possible to mix IDE hard drives with SCSI drives in a system.
Normally I dont like to do that because it can be very confusing to try to figure
things out. Simple is better. But keep in mind that drive mixing can be done.

Another interesting thing that you might get into, though its not as common today, is the need to know the number of cylinders and heads that an IDE
hard drive comes with. In computers with an older system BIOS, the computer
didnt recognize the IDE hard drive until you keyed in the number of cylinders
and heads the drive was using. Then the BIOS would (most times) recognize
the drive configuration and bless it as usable. Today the system BIOS autodetects the hard drives heads and cylinders and you dont have to go through
that rigmarole. The problem with the cylinders/heads scenario is that some
hard drives didnt come with that information stamped on them! You had to
go to a book or get on the Web (or on a BBS in the old days) to download a
schematic for the drive so you knew what to plug into the system BIOS.
SCSI is much easier to set up because you dont have to worry about getting
master/slave relationships right, nor do you have to be concerned about the
BIOS and whether it detected the drives heads and cylinders correctly. On top
of that, you can string several SCSI devices together (up to 7 for SCSI I, 14 for
SCSI II and III) so you can have a veritable Christmas tree of SCSI hard drives.
You have two or three issues to be concerned about with SCSI drives
though. First of all you need to be worried about properly cradling the drives
and getting adequate cooling to them. Its not a wise idea to cram bunches
of SCSI hard drives into a clone towers drive bay just because itll accept
them. Please be cognizant of the heat that a hard drive can put out and the
potential for burning up all hard drives in the system if you dont account for
cooling.
Also, youll have to make sure your SCSI IDs are correct. This is usually
quite easy to do. Most internal SCSI hard drives use jumper pins and youll
simply have to read your drives documentation to tell how to set it to the SCSI
ID youre interested in using. Typically you wont use ID 7. Thats most often
reserved for the SCSI adapter itself, hence the seven-drive SCSI I limitation.
I like to set it up so that in, say, a three-drive system, I set my boot disk
for ID 0, and the next two for ID 1 and ID 2. If you have a SCSI CD-ROM
youre hanging off the system (a pretty rare occurrence), you could set it at
ID 3. Ditto for other SCSI gear.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

74

Chapter 2

IDE Devices

Finally, its important to match the speed of the drives. Older SCSI drives
operate at 7500 RPM but todays SCSI drives run at 10,000 RPM. Its not
a wise idea to hang a 10,000 RPM drive in a system with other 7500 RPM
drives. I dont think itll break anything, but youll see variations in I/O and
could experience some funny activity with the machine.

Confirming Termination and Cabling

There is an incidental item that youll want to pay attention to with SCSI
the termination. Most of todays SCSI cards are self-terminating but youll
still have to consult your drives documentation to see if theres a termination
jumper pin. The basic rule is that either end of the SCSI chain is terminated.
So if your card is sitting at ID 7 and its terminated, youll also have to ensure
that ID 0 is terminated, provided, that is, youve got a hard drive hanging off
of ID 0. Note that its not necessary for you to start with ID 0.

Some older cards search the hard drives counting down in SCSI ID order.
Thus, in a configuration with a hard drive at ID 0 and one at ID 4, the system
would be trying to boot to ID 4 first. This can be, as you might imagine, very
confusing. Future Domain, a SCSI card company that was purchased by
Adaptec, operated this way. Watch out for this unusual behavior!

Youll need to be cautious of SCSI I, SCSI II, and SCSI III relative to the
cabling youll have to do both internally and externally. If youve got a SCSI III
adapter in the computer but the hard drive youre trying to connect to is SCSI I,
then youll need a cable that either has an adapter on one side or is SCSI Ito
SCSI III in design. This cable rule holds true for external devices connecting to
the external SCSI port as well. You can buy cables that are specially matched like
this, or you can simply buy an adapter. It might be a good idea to shy away from
adapters if you can, though in some circumstances you may not be able to.

Cabling and Master/Slave Relationships on ATA/IDE Drives

With IDE systems you generally have a master/slave relationship. EIDE can
support four IDE devices, so you can get into multiple-master/multipleslave relationships, but lets keep it simple for now. If you know that the
system is going to have one IDE hard drive and one IDE CD-ROM, its a
simple decision to make the hard drive master and the CD-ROM slave.
You want the system to boot to the hard drive and it wont do that if its
the slave and the CD-ROMs the master. Conversely, if you were to put

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Adding Hard Drives

75

two IDE hard drives into a computer, youd still be faced with making one
hard drive the master, one the slave. Generally, in situations such as this,
the OS will live on the master hard drive and the second hard drive will be
used for data. Both of the above IDE configurations are quite common.
What happens if you need two IDE hard drives, an IDE CD-ROM, and an
IDE CD writer? Well, then youre stuck with buying a second IDE controller
card or opting for an EIDE scenario. Todays motherboards typically include
IDE connections right on the board. No matter how you connect your hardware, one device will be master, one will be slave.
When setting master/slave relationships, youll almost always have to
adjust a jumper pin on the drive itself. These are clearly labeled. Read your
drives documentation.

Integrating the New Drives into the Storage System

Most of todays servers that are from tier 1 vendors come with a standard
drive cage that is designed to hold up to six drives. You can get servers that
have additional drive cages as well. The drive cage is connected by a cable
to the motherboards on-board SCSI adapter or to a RAID array controller
card. Typically the RAID cards are purchased separately but theyre well
worth adding to your initial server configuration. Youll purchase a certain
number of drives with the server and they will come equipped with connectors that allow them to slide into the backplane of the drive cage. Its a
very slick setup, allowing administrators to quickly change out hard drives
by simply sliding out one drive and sliding in a new one. In some cases you
may have to be cognizant of the SCSI IDs on the drives, but some newer
systems have gotten so fancy that the ID is automatically assigned based
upon location in the drive cage. Any vacant slots for which you didnt purchase a hard drive come with a blank so things look nice and neat.
Its key that you keep track of SCSI IDs so that if you decide to add
another SCSI device to the system, as with an external peripheral, you dont
occupy more IDs than are currently in use on the system.
Generally, in small servers that will have only one drive in them (as in a Windows 2000 Domain Controller [DC] box), youll want to purchase the RAID
card and two drives, then mirror the OS on the RAID card for added fault tolerance. But I think Id take it a step further and purchase a third drive to live in
the drive cage as a hot spare. If one of the hard drives fails, the computer wont
go down because its on a mirror. You can simply Plug and Play the hot spare
into place and rebuild the mirror. Different manufacturers have different methods for setting up hot spares, so read your system documentation.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

76

Chapter 2

IDE Devices

Summary

Between Chapter 1 and Chapter 2, you should have your disk subsystems
covered. If you are asked which is right for your implementation, you will have
a lot to think about, but you should be able to make an informed decision. As
far as the exam goes, make sure you are able to keep each of the ATA specifications straight. Fortunately, they make it relatively easy for you, just with the
naming convention. You should pay attention to things like when the 40/80
cable came into being, and how to choose a master or a slave device.
This chapter describes adding or changing hard drives in a system. You
essentially have two flavors of drives to consider: ATA/IDE or SCSI. There are
constant improvements and upgrades to each category and so you might wind
up changing out a SCSI I drive to a SCSI III that will result in I/O performance
enhancement. Telling the two types apart is easylook at the end of the drive
and count the pins. ATA/IDE is a 40-pin setup, SCSI varies from 50- to 68-pin
depending on the type of SCSI. Youll need to be aware of cabling issues with
SCSIsome installations may require a SCSI ItoSCSI III cable, for example.
Youll also have to keep in mind termination issues with SCSI. Generally
the SCSI adapter (ID 7) will be terminated and you may have to terminate the
other end of the chain as well. External devices use an external terminator,
while internal devices use jumpers for termination.
Modern servers use drive cages and ready-made slide-in devices that allow
for easy removal and upgrade of drives. These slide-in devices are proprietary,
so if youve got a slider for a Compaq computer it probably wont work in a
Dell and vice-versa.

Exam Essentials
Know that ATA and IDE are synonymous ATA is the official standard
defined term for IDE devices, though you will usually hear these devices
referred to as just IDE.
Know that IDE devices have the controller integrated into the drive
Unlike SCSI devices, which use an actual controller, the IDE controlling
device is contained as part of the drive. That is why it is referred to as
integrated.
Know the characteristics of each type of ATA device ATA was the first
type of IDE device, and was very limited in speed, addressable hard drive
size, and number of devices on the IDE chain. ATA-2 used DMA channels

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Summary

77

and PIO to increase speed, and it also introduced moving information

using a block transfer mode. ATA-3 does not increase speed, but it does
introduce security measures and SMART. ATA 33 increased speed and
used DMA and bus mastering. (ATA 33 is also called UDMA 33.) ATA
66 increases speed, uses bus mastering, and also changes the ATA cabling;
it is backwardly compatible with previous ATA devices. To take full
advantage of the ATA 66 devices, you need an appropriate cable and
controller. (ATA 66 is also called UDMA 66.)
Know the characteristics of Cable Select Cable select will determine
which drive is the master and which drive is the slave. It will assign the
master the designation C: and the slave the designation D:
Know why jumper settings are used in ATA devices Jumpers are used
to define devices as either Masters or Slaves. Therefore, there can be three
settings: Master/No Slave Present, Master/Slave Present, or Slave
Know how DMA works DMA is direct memory access. Instead of
requiring information to pass through the processor, the device can write
the information directly to memory. DMA predates bus mastering.
Know what bus mastering is and how it works Bus mastering is a more
efficient use of data flow. It is a program that is either in the microprocessor or in the I/O controller, which pushes information directly onto the
computer bus input/output paths. Once this is in place, data will go
directly from the I/O device to the processor.
Be able to install hard drives Know and understand the differences in
hard drives, cabling, termination, SCSI IDs, and IDE master/slave relationships and be able to explain how to add or upgrade hard drives in a
computer.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

78

Chapter 2

IDE Devices

Key Terms
Before you take the exam, be certain you are familiar with the following terms:
AT Attachment (ATA)
ATA Packet Interface (ATAPI)
ATA 100
ATA 2
ATA 3
ATA Packet Interface
block transfer mode
cable select (CSEL)
Cyclic Redundancy Check (CRC)
Direct Memory Access (DMA)
electromagnetic interference (EMI)
Enhanced IDE (EIDE)
Fast-ATA
Frame Check Sequence (FCS)
Identify Drive
Integrated Drive Electronics (IDE)
jumper
Logical Block Addressing (LBA)
master
Mean Time Between Failure (MTBF)
paddleboards
Parallel ATA
Programmed Input/Output (PIO)
Self-Monitoring Analysis and Report Technology (SMART)
Serial ATA
slave
Ultra DMA 33
Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

79

Review Questions
1. What does IDE stand for?
A. Integrated Device Electronics
B. Integrated DMA Efficiency
C. Integral Drive Economics
D. Integrated Drive Electronics
E. Interior Device Efficiency
2. What is ATA?
A. The name of an airline.
B. The actual standard that defines IDE.
C. IDE is the standard that defines ATA.
D. A type of burst mode DMA data transfer.
3. How does PIO work?
A. With PIO the input/output (I/O) goes directly to the processor.
B. With PIO the I/O is sent directly to memory.
C. With PIO the I/O bypasses the memory and the processor.
D. With PIO the I/O is sent simultaneously to the processor and to the

memory.
4. With DMA, where does the I/O go?
A. Directly to the processor.
B. Directly to memory.
C. The I/O bypasses the memory and the processor.
D. The I/O is sent simultaneously to the processor and to the memory.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

80

Chapter 2

IDE Devices

5. Which is the ATA specification where DMA is first used?

A. ATA
B. ATA 2
C. ATA 33
D. ATA 66
E. ATA 100
6. How many pins and conductors are in an ATA cable?
A. 20 pins 20 conductors
B. 20 pins 40 conductors
C. 40 pins 40 conductors
D. 40 pins 80 conductors
E. 80 pins 80 conductors
7. How many pins and conductors are in an ATA 66 cable?
A. 20 pins 20 conductors
B. 20 pins 40 conductors
C. 40 pins 40 conductors
D. 40 pins 80 conductors
E. 80 pins 80 conductors
8. In an ATA 66 cable, what are the extra 40 conductors used for?
A. Expansion
B. +5V
C. +.5 amps
D. Ground
E. Polarity
F. DMA

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

81

9. Monica is a network administrator who has been assigned the task of

upgrading the hard drives in an older server that has been running for two
years now. There is little documentation available for this server. What is
the first thing Monica must determine before she can go forward with her
hard drive replacement?
A. How many drives are in system
B. The type of hard drives
C. The SCSI IDs of all the hard drives
D. The master/slave relationship
E. What type of hard drives are in the computer
10. What is one of the reasons the ATA specifications are getting away

from the 40-pin, 80-connector cable?

A. The cable becomes brittle after several years and connectors break.
B. As data transfer rates increase, the signal inside the cable speeds

up, causing the electrons to go faster. This can cause the cable to
overheat, causing fires.
C. The cable can become twisted.
D. The cable can block airflow from the fan, causing excess heat

inside the case.

E. The specifications will stay the same.
11. How many devices can be in an ATA 66 chain?
A. 5
B. 4
C. 3
D. 2
E. 1

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

82

Chapter 2

IDE Devices

12. Which two of the following choices describe how devices that are part

of an IDE/ATA chain must be designated?

A. CD-ROM
B. Master
C. Optical drive
D. Tape drive
E. Slave
F. Target 1/Lun 1
13. What was one of the issues surrounding cable select?
A. There were none; it worked great first time, every time.
B. It would always designate the master device as Drive E:.
C. It would always assign the slave device as Drive D:.
D. It just didnt work as advertised.
14. What is the usual number of sets of pins available to designate master/

slave relationships?
A. There are usually two sets of three pins.
B. There are usually three sets of two pins.
C. There are normally six sets of two pins.
D. It varies.
15. What is the name of the device that connects the two pins to create an

I/O path?
A. Pin connector
B. Rocker switch
C. Bipolar DIP switch
D. Jumper

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

83

16. You have just been given a new IDE hard drive. When you check to see

if it is configured to be a master or a slave, you notice there are no

jumpers present. What can you derive from this observation?
A. The drive is a master device with no slave present.
B. The drive is a slave device.
C. The drive is a slave device with no master present.
D. The drive is a master device with a slave present.
E. Dont assume anything; check the documentation.
17. Youve recently changed out your servers IDE hard drive with a new

one but you cant seem to get the hard drive to come up and be recognized. There is an IDE CD-ROM in the system as well. What could be
the problem?
A. BIOS doesnt recognize the correct cylinders and heads.
B. CD-ROM is set to be master.
C. CD-ROM and hard drive are both set to be slave.
D. Termination jumper on hard drive isnt set.
18. What is the rated data throughput for an ATA 66 device?
A. 33 MBytes/second
B. 44 GBytes/second
C. 66 MBbytes/second
D. 66 KBytes/second
E. 66 GBytes/second
19. What is another name for an ATA 66 device?
A. Ultra 66
B. Supra 66
C. DMA 66
D. EMA 66

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

84

Chapter 2

IDE Devices

20. If there is a primary and a secondary ATA 66 controller built into a

motherboard, what is the maximum number of ATA devices that can

be put in the computer?
A. 2
B. 4
C. 6
D. 8
E. Unlimited

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Answers to Review Questions

85

Answers to Review Questions

1. A, D. Actually, depending on the data you gather, in some places it is

called Integrated Device Electronics and in others it is called Integrated Drive Electronics.
2. B. AT Attachment (ATA) is the actual standard that defines IDE.
3. A. With PIO, all I/O goes through the processor.
4. B. With DMA, I/O is sent directly to memory.
5. B. DMA is first used in the ATA 2 specification.
6. C. In the early ATA specifications, the cable had 40 pins and 80

conductors.
7. D. In an ATA 66 cable there are 40 pins and 80 conductors.
8. D. The additional 40 conductors are used for grounding to prevent the

introduction of noise and electromagnetic interference.

9. A, E. Monica must first determine how many hard drives shes replac-

ing and what type they areSCSI or IDE. Once she knows what type of
drive shes dealing with, she can ascertain their SCSI IDs or the master/
slave relationship. She should also ascertain the speed of the drives, if
SCSI. Most drive and schematic information is usually available on the
manufacturers Web site.
10. D. Because of the cable width, it can block airflow from the fan, caus-

ing the system to overheat.

11. D. There can only be two devices in an ATA 66 chain; however, there

can be two channels of two devices each.

12. B, E. If you have multiple drives in an ATA Channel, one must be des-

ignated as the master and one must be designated as the slave.

13. C. The slave device was always assigned as Drive D:.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

86

Chapter 2

IDE Devices

14. B. There are normally three sets of two pins.

15. D. It is called a jumper.
16. E. In 99.9% of the cases, the drive will be a master device with no

slave present. Just to be sure, check the documentation. It will save

you the hassle of taking the drive back out and re-jumpering it.
17. A, C. Typically, new IDE hard drives are set for slave, not master. This

is something youll routinely want to check when you purchase new

IDE hard drives. Also, on a computer with an older BIOS, you may
have to key in the amount of heads and cylinders that the hard drive
came with so that the BIOS can recognize the drive. Since youre
replacing the hard drive, chances are remote that the CD-ROM was
set for master. You dont have termination worries with IDEthats
a SCSI thing.+
18. C. The rated throughput of an ATA 66 device is 66 MBytes/second.
19. A. ATA 66 can also be referred to as Ultra 66.
20. B. You can have two devices on each controller for a maximum of

four devices.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Chapter

CPUs and Fibre Channel

SERVER+ EXAM OBJECTIVES COVERED IN
THIS CHAPTER:

3.2 Add Processors.


On single processor upgrade, verify compatibility.

Verify N 1 stepping.

Verify speed and cache matching.

Perform BIOS upgrade.

Perform OS upgrade to support multiprocessors.

Perform upgrade checklist, including: locate/obtain latest

test drivers, OS updates, software, etc.; review FAQs,
instruction, facts and issues; test and pilot; schedule
downtime; implement ESD best practices; confirm that
upgrade has been recognized; review and baseline;
document upgrade.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

n the last chapter, we spent a lot of time talking about how to link
physical hard disks together to give you more disk space and a sense of
redundancy in case of a failure. Now we are going to move from the disk
subsystem to the brains of the operationthe CPU and the ways that we can
maximize effectiveness.
As you look over the objectives, you will see a lot of attention paid to grouping CPUs together, either as part of the same physical computer with multiprocessing or by taking advantage of groups of servers by clustering. Clustering is
one of those buzzwords that just wont go away. It takes the concepts of RAID,
mirroring, and duplexing to a new height. Basically, we are moving the single
point of failure back from the disk subsystem, back even beyond the server. With
cluster servers, instead of having our data and applications protected by having
an additional disk subsystem, we are providing high availability of data and
applications by having additional servers.

For complete coverage of objective 3.2, please also see Chapters 6, 8, and 9.

Clustering

lthough clustering and cluster servers are the current buzzwords, the
concepts have been around for years. Actually, the implementations have been
around for years. The mainframe, big iron people had clustering almost since
day one, and on the LAN side, Novell had System Fault Tolerance systems
back in the days of NetWare 3. So, we are not talking about new technology.
When you start talking about clustering, you are actually opening up the
discussion of disaster recovery. Now, if you have ever participated in a disaster recovery exercise, you know that it can get to be pretty intense. When you

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Clustering

89

are planning for disaster recovery, the first thing you have to do is determine
how valuable your companys data is, and how long you can live without it.
Most members of senior management will tell you that the data is invaluable
and you cannot live without it even for a second. Then you start showing the
person how you can, in fact, ensure that data is always available with a
99.9999999% uptime, 24 hours a day, 7 days a week, 365 days a year. It is
an impressive display, until you get to the cost. That is when the rubber hits
the road.
So, what is clustering anyway, how does it work, and why is there the
potential for costs to skyrocket?

Clustering Basics
Clustering is basically having redundant, mirrored servers. In other words, if
one of the servers in your network were to fail over to its mirror, the other
server would immediately pick up the slack, and make all of the up-to-theminute data available to your users as well as all the applications that were
running on the failed server. In addition, for this to work really well, the
changeover should be transparent to the end user. In other words, Ursula
User would have no idea whether her requests for data and applications were
coming from Server A or Server B. Nor would she care.
So, now it comes time to determine what a disaster is and how can we protect against it using clustering, because after all, there are several different
kinds of disaster. Well, the first and most obvious example of a disaster is to
have something happen to the file server; lets say that someone was walking
through the computer room with a can of soda and tripped, spilling the soda
into the file server.

Now, okay, so this scenario may not be one that immediately jumps to mind.
But I have seen a file server that handled all services for a small law firm that
was physically located in the break room, in a small enclosure directly under
the coffee pot. Now, of all the places that I have seen servers placed, this was
the second most bizarre. The most bizarre was at a company that wanted to
prove to their customers how technologically advanced they were. To make
sure their customers could see the server screensaver when they walked into
the waiting room, that is where the server was placed. Now, if that were not
bad enough, the keyboard was attached and was active, as was the mouse.
So, anyone who came into the reception area that was really bored could
amuse herself by starting and stopping services or just rebooting the server.
We are not even going to mention the data that was available.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

90

Chapter 3

CPUs and Fibre Channel

Once the soda hits the file server, the smell of burnt silicon starts to permeate the building and that server is officially designated as toast. If this
were the disaster we were protecting against, our clustered server could be
mere feet away and still provide protection. In this case, just having a clustered server in the next room would be all the protection you would need.
Lets say there was a more serious problem. Suppose there was a fire in the
building that housed the file server. Now you can see that the only way clustering would work would be if the machine were physically located in a
different building, but the building could still be close by. If we moved the
disaster up in scale from impacting a single building to something like a flood,
tornado, or earthquake, now the clustered servers need to be several (or many)
miles apart to be safe. We can even take this a step further: Suppose you live
in a part of the world where political unrest is a way of life, or war is comm
onplace. In that case, you may want to have one of your clustered machines
located on the other side of the globe.
Clustering is just making sure that the mission critical business applications
and data that your enterprise requires to operate have high availability, meaning
that they are available 24 hours a day, 7 days a week, 52 weeks a year, year in
and year out. This high availability is usually necessary simply because of the
cost of operations. Lets say that you are talking about the application that runs
reservations for a major international airline. If that application is unavailable
for any reason, for any time, anywhere in the world, the loss of revenues to the
company could be in the millions-of-dollars-an-hour range.
Lets look at another example. Recently I read an article about the IS
department at the National Aeronautics and Space Administration (NASA).
It is responsible, among other things, for the computer network that tracks
the space shuttles when they are in orbit. This involves things like communication, tracking, navigation, life support, small things like that. Can you
imagine what the availability of that system must be every time a shuttle
takes off? I would imagine that having the space shuttle just take another
orbit while we reboot the server is not necessarily an option.

Clustering Technologies
Clustering offers differing challenges as you face each of the scenarios faced
above. Clustering, obviously, is a combination hardware and software solutions to the high availability challenge. This challenge may be something like
making sure that a database application is available no matter what the circumstances, or just making sure that a vital network service like e-mail is not
affected if one of the servers on the network should fail. Basically, a clustered
environment would look something like Figure 3.1.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Clustering

FIGURE 3.1

91

Cluster server environment

Active database server

Networked workstations
connected to the
database server

With cluster server, you have at least two servers, working together in tandem. If something were to happen to either of the servers, the other server
would be able to take over immediately. This means that if any of the server
applications were to fail, the cluster server software would restart any configured applications on any of the remaining servers. This seems to imply
that each of the cluster servers has to be configured exactly the same way,
and that is not necessarily the case. In some implementations you may have
two different applications running on the two servers, and if one server were
to fail, the other server would start the failed application and make it available. Look at Figure 3.2, which uses a database application and an e-mail
application as an example. This is the way the cluster would look before the
fail over.
FIGURE 3.2

Cluster server configuration before fail over

Active database server

Networked workstations
connected to cluster

Active e-mail server

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

92

Chapter 3

CPUs and Fibre Channel

Figure 3.3 shows the way the cluster would respond after a system failure.
FIGURE 3.3

Cluster server configuration after the fail over

Active database server

Access to database
and e-mail maintained

Active e-mail server

But providing access to applications is only half the problem. What about
providing access to the data that can be changed and updated on a minuteby-minute basis? Going back to our example of the reservation system for an
international airline, the application is not much use if the database for flight
and passenger information is not available. Therefore each server node in the
cluster must have access to the same information so that the application and
the data can be moved from one location to another without any downtime.
One of the ways this can be done is with shared external storage. Take a
look at Figure 3.4.
FIGURE 3.4

Shared external storage maintained with a SCSI bus

Cluster
server
nodes

Shared
SCSI bus

SCSI disks

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Clustering

93

In this case, both servers are accessing the same storage location, linked
together with a SCSI bus or a Fibre Channel configuration. It is up to the
cluster server software to decide which node has access to which pieces of
data at any given time. In this configuration, only one node can access any
information at any time. This is one way of making sure that the data on the
external storage is not corrupted.
The disadvantage of this configuration is that due to the limitations of
SCSI, the machines must be located close together. As you can see in Figure
3.5, shared SCSI technology has a distance limitation of just 82 feet.
FIGURE 3.5

Cluster server with SCSI limitations

Cluster
nodes

Shared
SCSI bus

Shared SCSI distance

limitation of 82 feet

SCSI disks

Even Fibre Channel technology has the limitation of several kilometers.

That would work well in our first scenario of someone spilling a soda on a
file server, but may not provide much protection against fire, flood, or earthquake. In this case, if someone in London called in looking for a plane reservation, the caller may have little or no sympathy that a backhoe in
Minneapolis cut a cable that provided data to the airline reservation system.
They would probably ask, and perhaps rightly so, why wasnt something like
this anticipated? Why should a relatively minor mistake in one part of the
world adversely affect computing in another part of the world?
So the challenge then becomes to protect not only the servers and their
applications, but also the data, and not in just a single location, but perhaps
worldwide. If you use clustered servers with RAID storage, with machines
with high-level uninterruptible power supplies, and even have emergency
power available, you appear to be taking precautions against most disasters,
but there are still those disasters that require more redundancy than even a
set of clustered servers can provide.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

94

Chapter 3

CPUs and Fibre Channel

There are solutions available that, for example, can use an IP network to
bypass the limitations of Fibre Channel or shared SCSI. When it comes time
to manage the data, it is handled like this. When there are changes to the data
on the primary node, these changes are captured and are sent via TCP/IP to
the backup node. That way, there is an exact copy of the data stored on the
second disk. If for any reason the primary data storage area should become
unavailable, the data is still accessible. In some cases, the solution can actually create multiple copies of the data, so even the backup is being backed up.
In this way, if there were a problem with the home site in Minneapolis, users
in different areas of the world would not suffer. Configured applications
would be back online within minutes and the data would be up-to-theminute. This would save tons of time over solutions like tape backups, where
the data is, at best, hours old, and at worst, days or weeks old.

Clustering Scalability
When it comes to scaling the clustering solution, you get what you pay for.
In some cases, you may only cluster on a one-to-one basis, so there is little
flexibility. With other solutions, you can configure the cluster to provide a
variety of solutions. Take a look at Figure 3.6.
FIGURE 3.6

Two-server basic cluster

Primary data server

Backup data server

TCP/IP network

In this case, we have the most basic clustering solution where one server
is acting as the primary and the other is acting as the backup. This is the
prime definition of clustering. Any data that is written to the primary is written to the backup. If something were to happen to the primary, the fail over
would bring the backup on line and life would go on with up-to-the-minute
data. In this case, there is one primary server and one backup server.
(I know the term Fibre Channel is new and we have not talked about it yet, but
we will, later in this chapter. Right now it is just important to realize that it can be
used to link servers with storage subsystems and it has a longer distance limitation
than SCSI. We will cover the rest of the stuff later!)

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Clustering

95

That is not necessarily the way it has to work. Using some software implementations, you can configure clustering so there are two primary servers and the data
replication is two-way, as shown in Figure 3.7. Now this configuration does have
a gotcha. In this case, the data has to be independent. Any data that originates on
one server can only be changed on that server. If it is changed on the backup server,
the changes will not be replicated back to the original server.
FIGURE 3.7

Two-way replication with both machines being the primary servers

One machine may be a dedicated

backup or both machines may
provide disaster recovery for the other
WAN

Now, there are other, more creative ways that you can use clustering solutions. For example, you can do what is called daisychaining clustered servers. In this case, lets say that we had some critical data in the office in the
Florida Keys. If the primary server went down, we wanted a rapid fail over,
so users could quickly pick up where they left off. That solution would
require a backup server on site, so we would not have to fight wide area network bottlenecks.
Because this data is critical and because we also understand that the Keys
are subject to hurricanes and other natural disasters, that could render the
two-servers-in-the-same-location solution worthless; we need to make
another backup copy off-site, somewhere far away. In this case, we can daisychain the servers, so there are two servers in the Keys, and another off-site,
away from potential storms and other disasters.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

96

Chapter 3

CPUs and Fibre Channel

Clustering Summary
Clustering is a viable solution, but the level of protection that you get
depends on the level of expenditure that you make. Some clustering solutions
that are right out of the box can handle only a one-to-one server relationship,
and even then, the servers have to be in close proximity. If you want true
disaster recovery capability where the servers are located hundreds of miles
apart, you are probably going to have to go with a specialty solution.

Real World Scenario

When most people think of clustering, they think of Microsofts clustering
solution. That is just one of the ways of accomplishing the task. There are
a variety of third-party vendors that offer clustering solutions for different
operating systems.
Clustering is called for when high availability is a way of life. If you have
servers on your network that absolutely, positively must be available or the
entire company might be at risk, clustering is the way to go. Make sure
when you do decide to cluster servers that you pay close attention to server
placement. Again, the purpose of clustering is high availability. If you place
both servers in close proximity to one another, you may be defeating the
purpose.
When you look at server location, be sure to look at the types of links that
are available between the locations. Clustering can be bandwidth intensive.
If you are going to be moving a lot of data between locations, make sure
that you have the infrastructure to support the demand. Also, you should
plan for the built-in latency of such a configuration. If you are crossing miles
of cable, obviously things are not going to be happening in real time. You
will need to build in solutions for this eventuality.
One of the ways that you can extend the area you cover is with Fibre
Channel. We will be looking at that next.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Fibre Channel

97

Fibre Channel

Now, one of the suggested ways that things be linked together is with
Fibre Channel. Lets take a look and see how that works, and what kinds of
things you can hook together.

History of Fibre Channel

The way things usually seem to work in the computer industry is that someone develops a technology and then after the new technology has been
around for a while, some group comes along and writes the standards for
that technology. If you remember back to Chapter 1, during our discussion
of SCSI that whole process was mentioned several times. Fibre Channel is a
little different. The standards were being developed long before the products
hit the market. As a matter of fact, the American National Standards Institute (ANSI) started working on the Fibre Channel standards back in 1989.
The whole point of the standard was to establish a high-speed connectivity
standard. Originally, as the name implies, this connectivity was to take place
over fiber optic lines. Now, if you look at data coming out of the Fibre Channel Industry Association (www.fibrechannel.org), they are quick to point
out that the channel can be either optical or copper.
One of the challenges that faced networking from the very beginning was
finding a way to move data quickly over long distances. It seems like there
has always been a need for high-speed communication to link host-to-storage or server-to-server. The applications, the processors, the storage, even
the workstations, seemed to get faster and faster, but the ability to get the
data from point A to point B didnt seem to keep up. Fibre Channel is a way
of addressing these needs. It is a gigabit interconnect technology that allows
multiple communications between workstations, mainframes, server, data
storage, and other devices all using either IP or SCSI technology. There are
ways to interconnect systems with multiple technologies to move data at
speeds that approach a terabit per second.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

98

Chapter 3

CPUs and Fibre Channel

Fibre Channel Terms

Yeah, I know, these will be defined later in the key terms, but you better have
a good grasp of them now, or the rest of this discussion probably wont make
much sense. These are some terms that are used a lot when talking about
Fibre Channel communication.
The first two to look at are channel and network. It is important to make the distinction. The channel is the point-to-point communication between two
devices. The point-to-point communication can be either a direct connection,
like between a server and an external disk subsystem, or it can be switched, like
a server-to-server connection using a switch. Channels are hardware intensive
and support the transportation of data at high speeds with low overheard. A
network, on the other hand, is the grouping of nodes that can communicate
based on a common protocol. So, when the user Brandice requests information from the disk subsystem, the request is carried over the network to the
server and the server accesses the information off of an external set of disks
using the point-to-point channel.
Next, there is the term Fabric. The Fabric is the active, intelligent interconnection scheme, or basically how these things connect. When you think of
Fabric, think about the part of the infrastructure that controls the routing.
Ports on the Fabric are called F_ports. You will also see the term FL_port,
which is just an F_port in an arbitrated loop environment.
The data flows between hardware entities called N_ports. The N_port is usually
a termination card that contains the hardware and software necessary to deal
with the Fibre Channel protocol. There must be at least one N_port in each
node, though usually there are two. You will also see the term NL_port, which
is just an N_port in an arbitrated loop environment.
The N_port has a unique address, called the N_port Identifier, and it also
contains a Link Control Facility (LCF).
If you can keep these terms straight, the rest of this section will be much easier.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Fibre Channel

99

Fibre Channel is not really one standardit is a set of standards. Take a

look at the following nested list of Fibre Channel Standards.
Fibre Channel Standards
ANSI X3T11 Fibre Channel Standards and Draft Standards
Fibre Channel Physical (FC-PH)
Fibre Channel Physical and Signaling Interface
Fibre Channel Reference Card
Fibre Channel-PH-2
Fibre Channel-PH-3
Fibre Channel Arbitrated Loop (FC-AL)
Fibre Channel Protocol for SCSI (FCP)
Fibre Channel Protocol for 802.2 Link Encapsulation (FC-LE)
Fibre Channel Protocol for High Performance Parallel Interface
(HIPPI) (FC-FP)
Fibre Channel Protocol for Single-Byte Channel Command Set
CONnection Architecture (SBCON)
Fibre Channel Generic Services (FC-GS)
Fibre Channel Enhanced Physical Interface (FC-EP)
Profiles
FCSI Profiles
FCSI Profile Structure (FCSI-001)
FCSI Common FC-PH Feature Sets (FCSI-101)
FCSI SCSI Profile (FCSI-201)
FCSI IP Profile (FCSI-202)
Gigabit Link Module Specifications (FCSI-301)
Loop Profiles
Private Loop Direct Attach Document
Public Look Profile (FC-PLP)

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

100

Chapter 3

CPUs and Fibre Channel

Other Fibre Channel Specifications and Documents

N-Port-to-F_Port Interoperability
10 Bit Interface Specification
Fiber Channel Management Information Bases (FC-MIBs)
Fibre Channel Optical Converter Proposed Specification

Fibre Channel Basics

Fibre Channel is a system where the designated ports log in to each other
through the Fabric. Given this design, the Fabric can be a circuit switch, it
can be an intelligent active hub, or it can be a loop. With Fibre Channel there
are three different topologies.

Point-to-Point
You remember point-to-point from back in the Network+ class, dont you?
This is the simplest of all topologies. With a point-to-point connection, there
is a bidirectional link that connects the N_ports on two nodes. A point-topoint topology will usually underutilize the bandwidth of the communications link.

Arbitrated Loop
With arbitrated loop, we start looking at a form of Fabric topology.


An arbitrated loop has shared bandwidth, meaning there are more

than two nodes connected.

It interconnects the NL_ports/FL_ports using unidirectional links.

There is only one active L_portL_port connection, so there is the

opportunity for blocking.

In order to guarantee access to the loop, there is a fairness algorithm in

place that ensures that no L_port is blocked from accessing the loop.

If any link in the loop should fail, the communication between all the
L_ports is terminated.

With an arbitrated loop, there is no switched Fabric, so it is less expensive

than Cross-point.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Fibre Channel

101

Cross-point or Fabric Switched

With a Cross-point or Fabric Switched network, you get higher performance
and connectivity.


There is a bidirectional connection between a node (N_port) and the

Fabric (F_port).

If multiple paths are configured between any two F-ports, it can be

configured as non-blocking.

This method has increased overhead with a destination identified

added to the frame header. This helps to route the frame through the
Fabric and get it to the desired N_port.

This topology provides the highest grade of performance and connectivity.

It efficiently shares the bandwidth.

Connectivity is guaranteed and there is no congestion.

If stations are added to the Fabric, it does not reduce the point-topoint Channel bandwidth.

Generic Fabric requirements are defined by standard (FC-FG).

Switch Fabric characteristics are defined by standard (FC-SW).

The topology of the Fibre Channel is completely transparent to the

attached nodes. The ports will all negotiate the type of connection, regardless of topology. There are some caveats though. For example, a point-topoint or an arbitrated loop topology requires that all nodes use the same data
transfer rate at all times. With Cross-point, there can be a dynamic rate conversion. With a switched topology, a 266Mbaud unit could connect to a
1.062 Gbaud unit. In this case, the ports would negotiate the highest shared
data rate for information transmission.

Using Fibre Channel

So far we have established that Fibre Channel is fast and that it has exceeded
the normal SCSI distance limits. Where may you see Fibre Channel in the
future?
One of the technologies that is starting to get more and more play is the technology of Storage Area Networks, where all the servers use one common storage
area for application and data storage. Obviously, in this case the storage would

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

102

Chapter 3

CPUs and Fibre Channel

have to be very-high-performance, which is a perfect implementation for Fibre

Channel. You will also see it in areas like these:


Very large databases or data warehouses

Backup subsystems and disaster recovery implementations like cluster

servers

High performance workgroups, like CAD/CAM implementations

Backbones for small, campus-wide networks

Networks that make use of digital audio and video

Storage Devices and Storage Area Networks

Fibre Channel is being provided now as a standard disk interface. As a matter of fact, controller manufacturers like Adaptec have several different
RAID controllers available that make use of Fibre Channel to the external
subsystem. When these RAID arrays are configured in massive sizes and
shared among servers, they become a Storage Area Network (SAN). To see
what I mean, check out Figure 3.8.
FIGURE 3.8

Diagram of a SAN

External SCSI disk

Server

Server

Server

Fibre Channel RAID

Switch or hub
Fibre Channel RAID

SCSI bridge
SCSI RAID

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Central Processing Units (CPUs)

103

Now the interesting thing about SANs is that both SCSI and IP protocols
are used to access the storage subsystems. The servers and the workstations
all use the Fibre Channel network to get access to the same sets of storage
devices or system. If there are older SCSI devices on the network, they can be
integrated into the Fibre Channel network through the use of the SCSI
bridge. What kind of performance are we talking about? Well, using a gigabit link, bandwidth is reported to be in the neighborhood of 97 MBytes/second for large file transfers.

Fibre Channel as a Backbone

Besides being used as a link to disk arrays or to cluster servers, Fibre Channel
can also be used as a backbone for campus networks. There are several reasons
why it works well in this implementation:


Confirmed delivery of packets.

Support for network resolution protocols like ARP, RARP, and others

Support for point-to-point configurations, shared bandwidth loop circuits,

and scalable bandwidth switched circuits.

Use of true connection services or services using fractional bandwidth.

These can be connection-oriented virtual circuits or real circuits.

Circuit setup time measured in microseconds.

Low latency connection-oriented or connectionless-oriented service.

Full support for time synchronous application like video.

Variable frame sizes.

Central Processing Units (CPUs)

he CPU is the brains of the server. It is responsible for the control and
direction of all the activities that the server participates in, using both the
internal and external buses. The CPU is just a processor chip that consists of
millions of transistors. That is what a CPU is and does. But like most things
in computing, there are dozens of processors.
When it comes to CPUs, there are only a few well-known manufacturers.
The best known, and the two manufacturers that are constantly battling it
out for the title of fastest, are Intel and Advanced Micro Devices (AMD). Of
the two, Intel is probably the more widely recognized, although AMD is
making inroads every day in the desktop and mobile computing market.
Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

104

Chapter 3

CPUs and Fibre Channel

Other manufacturers of server-quality processors include Motorola and

IBM, which have combined forces to create the Reduced Instruction Set
Code (RISC) chips. So, because we are talking about servers, we will limit
our discussions to Intel and RISC.

Back in 1965, one of the co-founders of Intel, Gordon Moore, was preparing
for a speech when he made a remarkable discovery. He discovered that the
number of transistors per square inch on integrated circuits had doubled
every 12 to 18 months since the processor was first invented. Moore speculated that the trend would continue, and that became Moores Law. If you
examine the prediction, you will find that it has been remarkably accurate. In
recent years the trend has slowed, and Moore has revamped the law to state
that the density of data will double every 18 months. Why are we mentioning
Moores law? Simpleeverything you are about to read about processors is
outdated. Most of it was probably outdated in the time it took this chapter to
go from my desk, through the editorial process, to the printer. So, if you are
reading this and thinking, What the heck is he talking aboutgigahertz is not
the state of the art, remember that when this was being written, the gigahertz
barrier had just been broken.

Intel Processors
At the time of this writing, the primary Intel processors on the server market
were the Pentium III and the Xeon Pentium III. Intel was also about to release the
first foray into the IA-64 architecture called the Itanium processor. Since the Itanium was scheduled to be the latest, greatest, bestest, fastest server processor on
the block, it was billed as the perfect solution for the large server market, even
before it was released. That left the Xeon and the Pentium III to hold down the
fort on the mid-range and low-end server market.

Intel also has the Celeron processor on shelves, but it was designed for the
lower end, desktop market. Because it is designed for the desktop, we wont
look at it here.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Central Processing Units (CPUs)

105

Xeon Pentium III

At the time of this writing, the Xeon was being sold in speeds of up to 1GHz.
The Xeon came in versions that supported two-way processing or multiprocessing. Two-way processing means having a motherboard with dual Xeon
processors, while multiprocessing can support two or more processors.
Other features of the Xeon include these:


Backwardly compatible with applications that were written for earlier

processors.

Optimized for 32-bit applications running on 32-bit operating systems.

Utilizes the Dynamic Independent Bus Architecture, which will separate

the dedicated external 100MHz system bus and the dedicated internal
cache bus operating at the full processor core speed.

Internal cache memory is a storage area that is designed to hold frequently-used

data and instructions. The processor contains an internal cache controller that
integrates the cache with the CPU. The controller cache does not store the actual
datait stores the RAM location of the actual data, thus providing faster execution of data and instruction sets. Because it is always faster to get information out
of memory than from anywhere else, the more cache there is, the faster the processor seems.

The Xeon can use 1MB or 2MB unified non-blocking, level-two cache.

100MHz system bus speed.

Can access 64GB of physical memory.

In the multiprocessing configuration, the Xeon can support four-way

symmetrical multiprocessing without specialized chipsets and clustering.
With specialized chipsets, it can scale to eight-way configurations.

Requires a mainboard specifically designed with the Xeon chipset.

More expensive than Pentium III.

For a more in-depth look at cache and at memory in general, read Chapter 4,
Memory.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

106

Chapter 3

CPUs and Fibre Channel

Figure 3.9 shows a picture of the Xeon Pentium III processor; the photo
was taken from Intels pressroom at http://www.intel.com/pressroom/
archive/photo/processors.htm.
FIGURE 3.9

Xeon Pentium III processor

Pentium III Processor

The Pentium III is an extension of the older Pentium II processors, which means
that a Pentium III would slip nicely into a mainboard designed for a Pentium II.
The only upgrade that would be necessary would be to upgrade the BIOS on the
mainboard to support Streaming Single-Instruction, Multiple-Data Extensions
(also referred to as SSE or Streaming SIMD). These instructions were designed
primarily for 3D graphics, popular in games.
When the Pentium III first came out, there was a bit of a controversy surrounding the chip. With the P-III, Intel promoted the concept of ID numbering and a random number generator that was based on the fluctuations of
heat within the processor. The plan was to make for more secure Internet
transactions, by turning the ID numbering off or on using software. The
instructions were really designed to make sure that the processors were not
being reworked so they could be overclocked. People concerned with privacy
took this new feature and became concerned that Internet sites would
require users to keep their ID turned on so they could access the site, thus
tracking a users movement over the Internet. The first set of P-IIIs had ID
tracking turned on by default.
The Pentium III has clock speeds of up to 1.13GHz, and it can use a
133MHz system bus.
Figure 3.10 shows a picture of a Pentium III processor.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Central Processing Units (CPUs)

FIGURE 3.10

107

Pentium III processor

The Pentium III is not as expensive as the Xeon processor, and the supporting
cast of mainboard and memory will bring down the cost also.

RISC Processors
You want power, we got power. Of course, like most things in computing,
the more performance you receive, the more you pay for it.
RISC chip servers are at the high end of the server platform, usually
reserved for the high availability, highly accessed Web servers. RISC based
servers can scale from a single processor up to 64 processors in the same
machine. Of course, the cost is going to be considerably higher than the usual
$10,000 to $15,000 price range for a starting server. In the case of a RISC
server, taking the cost well over $100,000 is not unheard of.
RISC is usually associated with Unix implementations, although Windows NT also ran on the RISC platforms.

Advantages of RISC
The RISC processor does offer several advantages over its Complex Instruction
Set Computing (CISC) counterparts:
Speed The name says it all. With RISC, you are dealing with a reduced
instruction set. That means that RISC processors often show two to four
times the performance of CISC processors in comparable technology and
using the same clock rates.
Simpler Hardware Because the instruction set is simpler, it uses up less
chip space. That means that extra functions like memory management or
floating-point arithmetic units can be installed on the same chip. Also,
since the chips are smaller, there can be more parts on a single silicon
wafer, and that reduces the cost per chip dramatically.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

108

Chapter 3

CPUs and Fibre Channel

Shorter design cycle Since the chips are simpler, it doesnt take as long to
design as the CISC brethren. This means that RISC chips can respond to
changes in the hardware marketplace sooner than the CISC designs. This
means there will be greater leaps in performance between the generations.

The Risks of RISC

Many of the problems associated with switching to RISC revolve around
software.
Code Quality The performance of the RISC processor depends on the
code it is executing. When the code is executing, it depends on something
called instruction scheduling to determine how quickly things work. If the
instructions are not scheduled correctly, the processor will spend lots of
time waiting for the result of one instruction before it can proceed to
another instruction. Therefore, the code written for RISC has to be written
and compiled in a high-level language like C or C++. Since C and C++ are
only languages, the quality of the compiler is what makes for efficient code.
Debugging If you have to worry about instruction scheduling, which is
normally an out-of-sight-out-of-mind process, debugging becomes difficult. If scheduling is turned off, the machine language instructions show
a clear connection with their lines of source code. Once scheduling is
turned on, the machine language instructions for one line of code may
show up in the middle of the instructions for another line of source code.
This not only makes the code hard to read, it makes it hard to debug.
Code Expansion Code expansion is where you take code that was originally designed to run on a CISC machine and recompile it to run on a
RISC machine. Since the RISC machine understands fewer instructions,
the same code may require more functions to accomplish the same tasks.
The exact size of the expansion depends on the quality of the compiler and
the machines instruction set.
System Design RISC machines run fast, and that means they have to
have really fast memory systems to feed them instructions. RISC-based
systems typically contain large memory cache, usually on the chip itself.
This is called first-level cache.

RISC Summary
While RISC is exceptionally scalable and works tremendously well in servers
that going to be heavily utilized, the monetary costs can be considerable.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Multiprocessing Support

109

Real World Scenario

The marketing from the motherboard manufacturers and from Intel will tell
you that Pentium III processors will work with motherboards designed for
Pentium IIs and vice versa, and that is true. But what they may not tell you is
that there are tricks involved in installing the processor.
If you buy a motherboard or a system that does not come with the processor, be sure you get all the documentation with the system. There are
switches or jumpers that may need to be set to make sure things match
between processor and motherboard. If they dont match, damage can
occur to the two most costly components of the system. It is truly a case of
RTFM, or Read The Fine Manual.
One of the things you should look for in the documentation is the motherboards beep codes. If there is something wrong during the Power On Self
Test (POST), the motherboard may not be able to display a message on the
monitor because it could be the video subsystem that is bad. Therefore, the
motherboard will announce errors through a series of beeps. At one time,
I worked for a tech support department of a computer manufacturing company, and one of the tests we used to run was to unplug everything from the
motherboard but the processor and the power and turn it on. When the test
was run, you expected to hear seven short beeps. If you didnt hear seven
short beeps, there was a problem with the motherboard.
If you install a processor on a mainboard, be sure the fan is installed properly and is functioning. Heat is a major cause of problems with processors,
and the first line of defense is the fan. Also, be sure the processor is properly
seated in its slot. Many times, new technicians are afraid to push down hard
on the processor because the motherboard may crack or break. Motherboards are more resilient than you may think.

Multiprocessing Support

or those of you who are fans of the American comedian Tim Allen,
perhaps we should just re-title this section More Power! See, unlike Allen, I
dont think it is just a guy thing. I want to make this more politically correct,
because everyone at one time or another wants more power. Certainly the
people on your network do, every time they complain about how slow the
network is running today. One of the ways that you can give them more
power is with Symmetrical Multiprocessing (SMP).
Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

110

Chapter 3

CPUs and Fibre Channel

You may be asking yourself, why all this fuss about multiple processors in
a single computer? After all, with the speed of processors getting faster all the
time, wont that take care of the issue?

Multiprocessing Basics
This will be like the discussion of RAID, complete with a whole new set of
acronyms and strange terms. Bear with me and it will make sense. First of all,
why might you need SMP?
When you take a look at the world of the uniprocessor (UP), you realize that
the processor is actually doing a lot of work at the same time. For example, the
processor may have a fixed-point arithmetic unit and a floating-point arithmetic
unit all on the same CPU. That means that the processor can run multiple
instructions within the same CPU. The thing to keep in mind is that while several
instructions can be run in parallel, only one task can be processed at a time.
Look at Figure 3.11.
FIGURE 3.11

Uniprocessing

Task

Task

Task

Task
Processor

In this case, you have multiple tasks backed up behind a single processor.
Now, you are probably saying to yourself, Wait a minute, he just said that
processors can perform multiple instructions at the same timewhat is the
difference? Think of it this way. Imagine yourself drying dishes after a big
meal. Each dish is a task. You may be able to dry multiple parts of the dish
at the same time, but you cannot dry multiple dishes at the same time. Adding another processor to the mix, like bringing in another person to help, will
cut the number of tasks down by half, and speed up the process of drying the
dishes. Figure 3.12 shows what I mean.
FIGURE 3.12

Multiprocessing

Task
Processor 3
Task

Task

Task

Task
Processor 2
Task
Processor 1

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Multiprocessing Support

111

Now, you would think that, like bringing in another person to help dry the
dishes, adding another processor would increase the overall performance of a
system in a directly proportional fashion. In other words, if you added a second processor to the system, the system would be twice as fast. It would be
wonderful if it worked that way, but it doesnt. You see, there a lot of other
factors that have to be taken into consideration. The problem is not just buying a motherboard that is compatible with two CPUs. All of the chipsets on the
motherboard have to be able to work with more than one CPU. The CPUs
themselves have to have hard-coded programming to work in parallel and,
once all the hardware is in place, the operating system has got to be able to
handle multiple processors. Now, all that has to happen to make sure these
two processors can work in tandem. Can you imagine how much behind-thescenes stuff has to go on to work with up to 64 processors? Not only that, but
there is still one more piece to the puzzle and that is the application.

Threadsthe Building Blocks of Tasks

A thread to the processor is the very smallest part of a task. For example,
when your alarm went off this morning, the reflex that caused you to move
your hand to shut off the alarm was a thread of that task. Now, if you are
looking for a more definitive explanation of a thread, it can be defined as a
concurrent process that is part of a larger process or program. In a multitasking operating system, a program may contain several threads, all running at the same time, inside the same program. This means that one part of
a program can be making a calculation while another part is drawing a graph
or a chartor making another calculation using another processor.
So, in order to use multiple processors, you have to have the ability to use
multiple threads, and these threads can be a piece of program, as long as that
piece does not depend on other pieces. In other words, when we use threads
in a multiple processor environment, each thread has to be totally focused on
its job. The thread can be split off and run either in serial or in parallel to
other threads without affecting its function.
There are two kinds of threads: kernel and user. The difference is what has
control of the thread and what is aware of the thread. In a kernel thread, it
is running close to the base operating system, so the kernel is aware of the
thread and has the actual control over it. With a user thread, on the other
hand, the process that spawns the thread has control over the thread. User
threads are usually faster to create and you are able to quickly switch contexts. User threads cannot be used across multiple processors, because the
kernel decides which CPU the thread is run on. If the kernel does not know
about the thread, then the thread can only run on one CPU.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

112

Chapter 3

CPUs and Fibre Channel

SMP Hardware
Obviously, when you are talking SMP, the hardware is important. With
some of the earlier Intel CPUs, you could mix and match older CPUs of close
clock speed, so you could do things like put a Pentium 166 in with a Pentium
200. You would just have to set both CPUs to run at either 200MHz or
166MHz, which of course could affect system stability.
Things are a little different with the more recent CPUs. With the more
recent systems, a multiplier is used. This multiplier is put into play to multiply the CPU bus clock rate, called the Front Side Bus (FSB) rate, by the multiplier. So, with a Pentium III 500, the FSB would be 100MHz and the
multiplier is 5, giving you the 500MHz. Intel now sets locks on the multipliers used in the CPU to control the final clock speed. Because of this, if you
are running multiple CPUs with Intel, you have to make sure the clock speed
and the multiplier are the same.
The CPU also has to have the onboard circuitry to work with other CPUs in
the same system. If that circuitry is not there, the CPU will simply not take
advantage of the other CPUs. This should not be a worry, because all of the Intel
CPUs that have been developed since the early Pentiums have had the ability to
work and play well in an SMP environment.

Intel does make the distinction between the dual processing (DP) environment and the multiprocessing environment (MP) for chips that are marked
with a VSU. If your Pentium chip has the VSU marking, it means the chip
has been validated to work in a uniprocessing and multiprocessing environment, but not in a dual processing environment. The difference is that DP is a
special mode of operation for two Pentium processors where there are four
dedicated private pins and there is the specific DP on-chip circuitry. This circuitry allows the processors to handle the negotiation of how to use the
resources and the data buses. Since there is no operating system intervention
required, this is referred to in Intel literature as a glueless solution. An MP
setup requires the glue like the operating system to negotiate between the
processors.

There are limitations to the number of processors that can be used. For
example, some Pentium IIIs can only be used in pairs, while the Pentium III
Xeon can be used in eight-CPU configurations.

Operating Systems
Having the hardware able to recognize the multiple CPUs is one half of the
battle, but the other half of the battle is the operating system. The operating
Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Multiprocessing Support

113

system has to be able to figure out that there is more than one CPU present
and also load the proper kernel. The kernel must then be multithreaded to
take advantage of the multiple CPUs inside the OS. This is really a bigger
issue than it may sound, because many of the system calls are static and cannot be reconfigured to work in a multithreaded environment. In that case,
some locks have to be put in place to make the system calls static.
The operating system is also responsible for system stability. It has to
manage the caches of the different CPUs. That management can get tricky,
because it has to make sure that the contents of the cache match each other,
as well as the original data, whether it is stored in RAM or on a disk. This
is one of the major hurdles for running multiple CPUs.
The OS must also support all of the processors that are available in the
hardware. An example would be that Windows 2000 Professional only supports 2 CPUs, so if you ran it on a system where there were four, two
wouldnt be used. Windows 2000 DataCenter Server supports 32 CPUs out
of the box. If you are using Linux, some of the Linux kernels natively support
16 CPUs, although the kernel code can be rewritten so more than 64 processors will be recognized.

Applications and SMP

Every application that runs on an SMP system has to be written or coded
very carefully to take advantage of these kernel threads whenever it can.
Since each application is different, using different sequences and using a different set of dependencies, the developer must figure out how to make the
best use of the multiple processors. When the developer has determined
those parts that can run independently of each other, these can be split into
separate chunks that are then perfect candidates for separate threads. For
those parts that arent independent, there must be some form of internal
locking mechanism to protect the calculations and the data they contain.
Because of the locking mechanism, there may be times that the developer
opts to not make use of the SMP. It may be that using another processor for
an independent piece would cause data corruption, meaning that it was more
trouble than it was worth.

You may also see the term processor or CPU affinity. This is the process of
selecting specific applications or processes to run on a specific processor. For
example, you may have a quad processor machine and want a database
indexing function to run specifically on the fourth processor. This would be a
function of affinity.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

114

Chapter 3

CPUs and Fibre Channel

SMP and Performance

When it comes down to SMP and performance, a lot can be summed up with
the phrase, It depends. For example, if you are using two processors on a
desktop machine that is running Windows 2000 Professional and you are
working with an application that is not calculation intensive, like Word, you
probably wont notice much of a performance increase. However, if you are
using some really math intensive applications like CAD/CAM, you may see
a remarkable increase.
The same is true with servers. The server will show peaks and valleys of
SMP performance gains. The server can use the multiple processors to offload its I/O processing across the CPUs. The difference is most noticeable
when there is heavy server usage. Extra CPUs and better handling of I/O will
make massive improvements in the performance when performance is
needed most.

Real World Scenario

I dont know about you, but one of the problems that I have is the "more is
always better" syndrome. You know what I mean: If 256MB of RAM is good,
512MB has to better. If one processor is good, two must be better, and four
must be better still. As a matter of fact, I have a friend who is running a mail
server in his office with four processors in it. He thinks it is really cool. The rest
of us think it may be a tad bit of overkill to run a mail program that services less
than 100 people. So, when it comes to processors, more being better may not
necessarily be the case. Before investing in a super server with 16 processors,
examine your alternatives and examine the applications that are going to run
on the server. You may find that the operating system, contrary to marketing
claims, may do a great job with multiprocessor support for two or four processors and may level off after that. You may increase your performance and get
better results (and it may even be cheaper) if you buy four servers with four
processors rather than one server with sixteen processors.
Do your homework and remember that more is not always better.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Adding Processors

115

Adding Processors

Adding processors can be a much less scary prospect if you do some basic
research on the server youre upgrading prior to attempting the addition.
There are several key things to keep in mind as you toy with the idea of
upgrading or adding new processors.

Verifying Compatibility in Single Processor Upgrades

Its extremely beneficial for you to read your product documentation or consult with the vendor or manufacturer regarding processor upgrades. The first
thing you need to ascertain is whether you can add multiple processors to your
computer. If so, how many can you add and what types will the bus accept?
If your server can accept only one processor, you must ascertain what types of
processors you can add. By types, I mean what brand of processor and what
speed? For example, if the computer currently has a Pentium II 200MHzz processor in it, can you upgrade to a Pentium III (answerno)? Can you upgrade
to a 450 or 500MHzz processor (answerprobably).
There is probably no greater criticality in server upgrades than being sure
to correctly identify and procure the processor upgrade that your computer
manufacturer has said will be a suitable upgrade for the system. If you dont
follow manufacturer guidelines and go with a clone, youll likely encounter
many problems.

Verify N 1 Stepping
In the world of CPU manufacturing, the word stepping is akin to a version
number. When a new microprocessor is released, the product version is set at
step A-0. Later on, as engineering updates are made to the chip, new steppings
are assigned. If the change is minute, the number of the stepping will be
changed (i.e., A-0 to A-1). If the change is major, the letter of the stepping will
change (i.e., A-0 to B-0).
When considering a CPU upgrade, especially if youre adding a CPU in
order to turn the system into a multiprocessor computer, youll want to verify
the current CPUs stepping and match accordingly, or replace if the stepping
levels are too far from one another. Check with the computer manufacturer or
vendor for more detailed compatibility information.
In a single CPU upgrade, the same caveats apply (matching the stepping
to the range supported by the manufacturer).

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

116

Chapter 3

CPUs and Fibre Channel

Verify Speed and Cache Matching

Additionally, in multiprocessor systems its important to match the existing
CPUs speed with the new addition to the family. This may be extremely difficult to do. If youve got a Pentium II 200MHz processor, you may well be out
of luck finding a match. You might have to upgrade the first processor and buy
a match for it in order to have matching CPUs. All of the CPU speeds in a multiprocessor system must match. You cannot have one CPU running at
133MHz, another at 700MHz, and so onit wont work.
Additionally, you must match the L2 cache on each processor. If the first
processor has a 250 L2 cache, its paramount that the additional processor
youre purchasing have the same size L2 cache.

Note that some processors require a DC power supply and have an associated
slot on the motherboard for the power supply unit. Note for sure whether your
processor implementation has this and order accordingly. If in doubt, ask the
manufacturer or consult documentation.

Summary

You know the problem with writing a chapter like this? As you write
about all the exceptional technology, you just want to go out and set up a
cluster of servers with four Xeon processors, a couple of gigs of RAM, and
a storage system hooked to the main box by Fibre Channel with a few
terabytes of disk space, just to see me if you can get it to work! Hmm, maybe
I could build it and sell it to my wifes company. Do you think that might be
a little overkill for a home-based business with 10 employees?
Anyway, enough of all this dreaming stuff. On to Chapter 4, where we
look at what kinds of memory to put into that big ol badboy.
You know, I could probably get all those servers to fit.
We talked about adding processors to a system. Its important to verify
your systems capabilities, either by checking with the manufacturer or consulting system documentation. Systems are typically rated for a given range
of microprocessors so you may not be able to run out and buy the latest and
greatest processor, slap it in your system, and hope that it works. Its important to understand your systems limitations.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Summary

117

Be sure to check the current processors stepping, L2 cache capacity, and

processor speed so that in a multiprocessor environment youre purchasing
like equipment. Dont mix L2 cache or processor speeds. Check with the
manufacturer to see what are allowable steppings between processors in a
multiprocessor system.

Exam Essentials
Know what it means to cluster servers Servers are clustered for a variety
of reasons, usually to make sure that the single point of failure is moved
back beyond the server. You can think of cluster servers as mirrored servers, though in reality, clustering can provide a broader range of services
than just fault tolerance.
Know what high availability means High availability is one of those
buzzwords that means exactly what it implies. You want your network to
be available always, 24 hours a day, 7 days a week. You take all the steps
necessary to make sure your server is up and running to provide the
appropriate services and applications to your users. It is highly available.
Know the basics of Fibre Channel Fibre Channel can be used to link
storage subsystems (or other devices) to the network. It provides faster
throughput. Fibre Channel makes use of ports connecting using the Fibre
Channel fabric. Fibre Channel is used in storage area networks, to provide
the bandwidth for remote access to large databases, and to provide bandwidth for remote backups.
Know about the different types of CPUS, including RISC, Pentium II,
Pentium III, and Xeon Xeon supports two-way processing and multiprocessing. Xeon supports four-way multiprocessing without specialized
chipsets. They are more expensive than the Pentium III processors.
Know which CPU you would use in a high availability super server
The RISC processor and the Xeon are designed for high availability and
high utilization servers.
Know the advantages and disadvantages of multiprocessing support
Before adding multiple processors, it is best to do a cost analysis. In some
cases, it may be cheaper to add another server with fewer processors than
it is to add a mainboard that can support more processors. For example,
it may be cheaper to provide a cluster of four servers with two processors
each than a single server with eight processors.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

118

Chapter 3

CPUs and Fibre Channel

Make a checklist Know and understand the things to check for when
upgrading a system processor or adding processors to a multiprocessor
system.

Key Terms
Before you take the exam, be certain you are familiar with the following terms:
American National Standards Institute (ANSI)
arbitrated loop
channel
cluster servers
Complex Instruction Set Computing (CISC)
Cross-point
dual processing (DP)
F_port
FL_port
Fabric
Fabric Switched
Fibre Channel
Front Side Bus (FSB)
gigabit
kernel thread
Link Control Facility (LCF)
N_port
NL_port
point-to-point
Reduced Instruction Set Code (RISC)
stepping
Storage Area Network (SAN)
Streaming Single-Instruction, Multiple-Data Extensions

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Summary

Symmetrical Multiprocessing (SMP)

terabit
thread
uniprocessor (UP)
user thread
Xeon

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

119

120

Chapter 3

CPUs and Fibre Channel

Review Questions
1. When servers are clustered, you are providing redundancy of which

devices?
A. Network cards
B. Mainboards
C. RAID systems
D. Servers
E. Video cards
2. What are the key items that must match when youre attempting to

add another processor to a multiprocessor system?

A. L2 cache
B. Cooling fan
C. Speed
D. Stepping
E. Model
3. Youre working on a server replacement for a two-year-old computer

thats had better days. You need to verify the CPU stepping. How do
you go about gathering this information?
A. Read the serial number on the CPU and call the manufacturer.
B. Read the stepping number on the CPU.
C. See if the NOS reports the stepping number.
D. Obtain the stepping number from the Web.
E. Read the system documentation.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

121

4. Which Intel processor is designed for higher scalability?

A. Celeron
B. Pentium II
C. Pentium III
D. Xeon
E. RISC
5. Louis has recently purchased and added a second processor to his

server. When the server boots, only one CPU reports online and the
NOS error logs report something about an L2 problem. What does
Louis need to check?
A. Secondary cache is mismatched between the two processors.
B. BIOS version is different between the two processors.
C. CPU speed is different between the two processors.
D. DC power converter missing.
6. The Linux 2.2 kernel can use up to 64 processors if what is done?
A. The kernel is rewritten and tweaked.
B. Nothing.
C. The processors are set to operate in parallel mode.
D. The processors have VPU on them.
7. What is the minimum number of servers in a cluster?
A. 4
B. 3
C. 2
D. 1

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

122

Chapter 3

CPUs and Fibre Channel

8. In a cluster server environment, there can only be one primary server?

A. True
B. False
9. RISC chips were designed by what two companies?
A. HP
B. Unix
C. Sun
D. IBM
E. Microsoft
F. Motorola
G. Novell
10. RISC stands for which of the following:
A. Reduced Instruction Set Computing
B. Redundant Installation Sun Coprocessors
C. Redolent Instructions to Sun Computers
11. CISC stands for which of the following:
A. Ciscos symbol on the New York Stock Exchange
B. Computerized Instruction Set Configuration
C. Complex Instruction Set Computing
12. What is one commonly overlooked checklist item to be sure you

include when considering a processor upgrade?

A. Matching of L2 cache
B. Verifying the need for additional DC power supply
C. Matching of CPU speed
D. Matching the CPU stepping number

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

123

13. Alejandro is trying to add a second processor to his server but the

computer wont boot. What could be the problem?

A. Mismatched L2 cache
B. Mismatched stepping number
C. Mismatched speed
D. No DC power supply for second CPU
14. What is the performance difference between RISC and CISC processors?
A. RISC processors often show two to four times the performance of

CISC processors in comparable technology and using the same

clock rates.
B. RISC processors often show 12 to 14 times the performance of

CISC processors in comparable technology and using the same

clock rates.
C. CISC processors often show two to four times the performance of

RISC processors in comparable technology and using the same

clock rates.
D. CISC processors often show 12 to 14 times the performance of

RISC processors in comparable technology and using the same

clock rates.
15. You have just mounted a Pentium III 500 processor on a mainboard

designed for a Pentium II. What else must you do?

A. Nothing.
B. Buy a new mainboard, because the Pentium III will burn out the

old one.
C. Buy a new processor, because the mainboard will burn out the

old one.
D. Flash the BIOS to provide for the extra instruction sets of the

Pentium III.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

124

Chapter 3

CPUs and Fibre Channel

16. When the Pentium III first came out, there was some controversy

surrounding the processor. Why?

A. It was rumored to be made out of parts from the endangered

species list.
B. The floating-point decimal was not always accurate.
C. It had ID tracking.
D. It was thought to gather an inventory of hardware and software on

your PC and e-mail it directly to Microsoft.

17. What are some considerations to take into account when youre add-

ing two more processors to a server that already has two, thus turning
it into a four-way computer?
A. Stepping of all four processors must match.
B. L2 cache of all four processors must match.
C. Speed of all four processors must match.
D. Must have ports available on motherboard for additional processors.
18. What is the minimum number of nodes in an arbitrated loop?
A. 4
B. 3
C. 2
D. 1
19. What agency certificated the specifications for Fibre Channel?
A. ASCII
B. SCSI
C. ANSI
D. EPSIDIC
E. IEEE

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

20. In a Fibre Channel configuration, where is the N_port located?

A. The N_port is usually a termination card.
B. The N_port is usually a main port.
C. The N_port is usually a virtual connection.
D. The N_port is usually a randomly defined memory address.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

125

126

Chapter 3

CPUs and Fibre Channel

Answers to Review Questions

1. D. Cluster servers means that two or more servers are acting in tandem

to provide data and applications in case of a failure.

2. A, C, E. Its important to match the CPUs brand, model, L2 cache, and

speed. The stepping may be importantcheck with the manufacturer.

The cooling fan probably wont matter.
3. A, B, C, D. Reading the system documentation isnt likely to provide

any information on the stepping number of the CPU. Remember that

a stepping number is like a version number and a documentation
booklet isnt likely to give you that kind of information. Your quickest
bet is to see if the NOS reports the stepping numbersome do and
some dont. With some manufacturers, you can go directly to their
Web site, key in the unit number for your computer, and get exact
details for the unitwhich may include the CPU stepping number. Its
doubtful that the CPU itself will include the stepping number, but if
you have to power the box down to get its serial number anyway, you
can take a look to see if its listed.
4. D. The key here is that the question specifically asks for Intel processors.

In that case, the answer would be the Xeon processor.

5. A. Secondary cache errors most likely mean that Louis has got an L2

cache mismatch. L2 typically comes in sequences of 256 so he may

have one CPU with 256K of cache while the new one has an L2 cache
of 512.
6. A. The Linux 2.2 kernel can use up to 64 processors if the kernel is

rewritten.
7. C. You must have at least two servers in a cluster.
8. B. You can have two primary servers. It is important to note that the

data will not be synchronized between servers.

9. D, F. RISC processors were designed by IBM and Motorola.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Answers to Review Questions

127

10. A. RISC stands for reduced Instruction Set Computing.

11. C. CISC stands for Complex Instruction Set Computing.
12. B. All of the items are important things to consider but an often-

overlooked item is the DC power supply. Generally manufacturers

will supply this item as a part of the equipment, but its important to
validate first that you indeed need one and second that youll get an
additional one with your order.
13. C. If the CPU speed is mismatched, the system wont operate.
14. A. RISC processors often show two to four times the performance

of CISC processors in comparable technology and using the same

clock rates.
15. D. The only upgrade that would be necessary would be to upgrade the

BIOS on the mainboard to support Streaming Single-Instruction,

Multiple-Data Extensions.
16. C. The Pentium III had ID tracking, which people felt could cause an

invasion of privacy.
17. B, C, D. The stepping number isnt nearly as important as the speed

and L2 cache of the processors. Check manufacturer documentation

for compatibility rules. The motherboard must be able to accept the
additional CPUs.
18. B. An arbitrated loop has shared bandwidth, meaning there are more

than two nodes connected.

19. C. Fibre Channel is defined by the ANSI X3T11 Fibre Channel Standards

and Draft Standards.

20. A. The N_port is usually a termination card.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Chapter

Memory
SERVER+ EXAM OBJECTIVES COVERED IN
THIS CHAPTER:

3.4 Increase memory.


Verify hardware and OS support for capacity increase.

Verify memory is on hardware/vendor compatibility list.

Verify memory compatibility (e.g., speed, brand, capacity,

EDO, ECC/non-ECC, SDRAM/RDRAM).

Perform upgrade checklist including: locate and obtain latest

test drivers, OS updates, software, etc.; review FAQs,
instructions, facts and issues; test and pilot; schedule
downtime; implement using ESD best practices; confirm that
the upgrade has been recognized; review and baseline;
document the upgrade.

Verify that server and OS recognize the added memory.

Perform server optimization to make use of additional RAM.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

n Chapter 3 we mentioned things like cache memory, but we

really have not had the opportunity to delve deeply into the subject. Now, if
you are one of those people who believe that memory is memory is memory;
I hate to be the one to burst your bubble, but wrongo! I am going to be
throwing around terms and acronyms like EDO, ECC, DIMM, SIMM, L1,
L2, RAID cache, writing back and writing through (also known as writing
thru), and a whole lot more.
In this chapter we are going to be looking at everything you need to know
about cache memory and how to configure memory hardware for optimum
server performance.

For complete coverage of objective 3.4, please also see Chapter 9.

Memory Types

ou have to remember where I am coming from. The very first computer that I ever bought came standard with 512K (thats right, K) of Random Access Memory (RAM). Now, I will never forget the look on the
salesmans face when I told him that I wanted to upgrade my system to 1
megabyte (MB) of RAM. He thought I was nuts. He actually told me that
I was throwing away my money, because there would never be a use for that
much memory. Now, some operating systems have minimum suggested
requirements of 128MB for installation. Guess my friendly computer salesperson was wrong!

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Memory Types

131

So, why is memory so important? If you want to speed up the performance of any PC or server, one of the first things you can do to it is add more
memory. As a matter of fact, that is a pretty common solution to server problems. It is always easier for a Central Processing Unit (CPU) to grab information out of memory than it is for the CPU to have to go look for it on a
hard disk, or in its instruction set. So, the more of the commonly referenced
information we can store in memory, the quicker the CPU can find it. The
faster the CPU does its job, the faster the server (or even a workstation)
appears. It is as simple as that. How does the CPU know what is commonly
referenced information and what isnt? It doesnt. So it just stores as much of
the stuff that people have asked for as it can. When people (or the system)
ask for stuff that it does not have in memory, it will usually rid itself of old
stuff that no one has asked for in a while and replace it with the new stuff
people (or the system) have recently asked for.

DIPs
Memory comes in various shapes and sizes, so lets start by taking a look at
some of the physical types of memory. I mentioned above that I had to
upgrade my first computer from 512K to 1MB of RAM. This involved a
technician adding some integrated circuits called dual inline packages
(DIPs) to the mainboard. These types of DIPs are shown in Figure 4.1. DIPs
have come in a variety of sizes, but now they are usually at least 256K per
DIP. To be honest, I have no idea what they were when I bought my first
computer, because for the first year I owned it, I was afraid to take the top
off for fear all the electrons would escape.
FIGURE 4.1

DIP memory chip

As the need for memory began to increase, a decision had to be made.

These DIPs plugged directly into the mainboard, and with four or eight of
them per megabyte, they began to take up some serious room. So, the choice
was to increase the size of the mainboard or find a different configuration for
memory. The designers opted for the different configuration of memory,
called the Single Inline Memory Module (SIMM).

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

132

Chapter 4

Memory

DIPs are still used for a variety of memory. For example, VGA cards or network
cards that have onboard cache will normally use DIPs.

SIMMs
The SIMM was just a different configuration of the DIP. Two types of SIMM
are shown in Figure 4.2.
FIGURE 4.2

SIMM memory

SIMMs were a breakthrough at the time of their introduction. The first

SIMMs had nine small DIP chips on them and took up less room than DIPs.
As a matter of fact, four SIMMs could be installed in the same space that
used to be allocated to a row of DIP chips. Installing SIMMs was relatively
easy: You placed the module in a slot at a 45-degree angle and gently pushed
down and rocked it back until it locked in place. There were some tricks with
SIMMs, the main one being that it always paid to buy the same type of
SIMM from the same manufacturer. The original SIMMs didnt work and
play well with SIMMs manufactured by different companies.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Memory Types

133

SIMM installation was actually a little more difficult than it sounds. Many of
the mainboards had plastic connectors, and if you were not careful, you could
break off the plastic. When that happened, the SIMM was not held securely in
its slot and it did not work well. This was usually time for a new mainboard,
and those were always expensive. For a while, I worked as a telephone technical support person, and my job was to talk people through the installation
of memory SIMMs. As a technician, I always warned the installer to be really
careful, and I just hated it when I heard something like, Oh darn, look what I
did coming out of the phone.

Getting back to Figure 4.2, you can see that the SIMM, depending on age,
comes in two different configurations. There was the 30-pin configuration
and the 72-pin configuration. When the 30-pin SIMMs first came out, computers were working with 32 data bits. Unfortunately, each SIMM only handled 8 data bits, so you needed to provide one bank of four SIMMs. A
memory bank was simply a set of four slots. Most computers had two banks
of four SIMMs available, Bank 0 and Bank 1. The CPU would then address,
or work with, one memory bank at a time.
72-pin SIMMS took care of part of the problem, because each 72-pin
SIMM supported 32 data bits. If you were using a 486 CPU from Intel or a
68040 from Motorola, you only needed one 72-pin SIMM per bank to give
the CPU the 32 data bits it was looking for.

Working with the early computers was always fun, because they never ceased
to provide unique opportunities. One of the opportunities was something
called chip creep. If you remember all the way back to high school science,
when things heat up, they expand; when they cool down, they contract. The
same is true with chips. After a computer had been turned on and off several
dozen times, the chips, which had expanded and contracted several dozen
times, may have worked themselves just ever-so-slightly out of their slots.
That meant the chip was not making proper contact and the thing didnt work
as advertised. As a user, you became adept at taking the top off your computer and gently pushing down on all the chips to reseat them.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

134

Chapter 4

Memory

DIMMs
After the SIMM came the Dual Inline Memory Module (DIMM). Look at
Figure 4.3.
FIGURE 4.3

Two types of DIMM

SO DIMM

168-pin DIMM

As you can see, there are two types of DIMM, but most of them installed
vertically into the mainboard, just like the SIMM. The difference between
SIMMs and DIMMs is in the pin configuration. On a SIMM, the opposing
pins on either side of the board are tied together to form a single electrical
contact. With a DIMM, the opposing pins remain separate and isolated to
form two contacts. DIMMs therefore usually have memory chips on both
sides of the module. DIMMs are used in 64-bit computer configurations.
This relates to the Intel Pentium or the IBM RISC processor.
At the top of Figure 4.3 is the Small Outline DIMM or the SO DIMM.
This DIMM is like a 72-pin SIMM in a reduced size. It is designed primarily
for laptop computers.
Next is the 168-pin DIMM. If you look carefully at it, you will notice the
notches in each side of the module. Instead of having to install this module
by inserting it at a 45-degree angle and rocking it back, this module slides
into its slot with rocker arms on each side. You start the installation by opening the rocker arms, and when you push the DIMM into the slot, the rocker

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Cache Memory

135

arms close and lock the module down. The rocker arm will then hold the
module firmly in place, eliminating any chip creep.
Now that we know what memory physically looks like, lets see how it
is used.

Cache Memory

For your basic server, there are two types of memory: cache memory and
main memory. Main memory is referred to by a variety of names, including
Dynamic Random Access Memory (DRAM) or just plain ol RAM. DRAM is
the part of memory that is responsible for holding instructions and for holding
data that will be used by the applications running on your server. It is also used
by the server operating system itself. When the servers CPU executes an instruction from an application, it goes out to RAM to see if there is information stored
in memory that it can use. DRAM is kind of the holding area for information
that may be accessed in the near future. Depending on the server, the amount of
DRAM can measure in the gigabytes.
There is another type of RAM, called Static Random Access Memory
(SRAM). Your first question is probably, Wait a minute. How can it be
static and random at the same time? Good question! SRAM is called static
because the information doesnt need to be updated very often. With memory, this update process is called a refresh. SRAM is usually physically bulky
and limited in its capacity. SRAM usually comes in a DIP. SRAM can be used
for cache. Lets start looking at cache and then we will explore the different
types of main memory.
Cache comes in much smaller amounts and it is much faster than main
memory. It is usually measured in the kilobyte range. The express purpose of
cache is to make it easier for the component to respond to request for services. Cache memory is used for the processor, it is used for RAID controllers, and it is even used for some types of network cards. In this section, we
are going to look at how a processor uses cache, how RAID uses cache, and
the differences between write-back and write-through cache. (You
might have noticed these quoted terms can also be spelled write-back and
write-thru. Either way, they mean the same things.)

Processor Cache
When a processor wants to access information, it wants that information as
quickly as it can get it, by using the fewest number of clock cycles. When you see
listings for the cache memory that will be used expressly for the processor, note
Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

136

Chapter 4

Memory

that it comes in either Level 1 cache (L1), or Level 2 cache (L2). Level 1 cache
is physically in the actual processor itself. Level 2 cache is usually part of the
mainboard and is dedicated strictly to providing memory for the processor.
Cache memory is always SRAM. SRAM can be on a DIP, a SIMM, or a
DIMM. The cache memory controller is the brains of the cache memory system. When the cache memory controller goes out to get an instruction from
the main memory, it will bring back the next several instructions and keep
them in cache, also. This happens because it is very likely that these instructions will also be needed. Because the instructions are already loaded in
memory, when the CPU makes the call for them, the instruction will be read
from cache, making the computer run faster. When the computer runs faster,
the user is happier and the network administrators life is easier.
So whenever you see the term cache, remember that this is just another
way to speed things up. Any time something can be read from memory,
rather than having to go to the hard disk or to the BIOS to find the information, it is going to take less time. Cache is just a segment of memory that has
been reserved by the component involved to temporarily store information
or instructions for faster retrieval.
Now, another place where cache memory is used is in RAID systems.

RAID Cache
RAID cache is a perfect server implementation for cache to come into play.
Think about it. You have several hundred people trying to access information from a RAID system or write information to the subsystem, almost
simultaneously. Now, I dont know about the users you have dealt with, but
the users on my network never understood the word patience. The biggest
complaint, it seemed, that people had was the speed of the network. Certainly, those few extra seconds were going to materially affect the standard
of living of some of these people!
Anyway, assume that we are looking at a busy server, or a server that is
dealing with lots of small I/O reads and writes. In this case, if the RAID controller were to become overwhelmed with work, it might have to put some
of the requests on hold while catching up. This is usually not a good thing.
So, the RAID controller uses cache as sort of a waiting room for requests. If
it cannot answer the request immediately, it may place the request in cache
until it can get to it.
As you imagine the differences between L1, L2, and RAID cache, notice
that one of the biggest differences is in size. L1 and L2 are measured in kilobytes while RAID cache is measured in megabytes. As a matter of fact, several
RAID controllers have minimum sizes before the cacheing will kick into effect.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Cache Memory

137

Write-Back vs Write-Through Cache

Here we go back to the CPU again. Since we are discussing writes here, we
are looking at how information is being written to the cache memory the
CPU actually uses. There are two ways this can happen, with write-through
or by using write-back. Here is how the two differ:


With write-through memory, each time there is a write operation to

the cache, it is accompanied by an operation that writes the same data
to main memory. If you have a system that is using write-through
cache, then an I/O processor does not have to look in the cache directory when it reads memory. After all, everything is in main memory
and main memory is usually faster than SRAM. This method makes
the access for the I/O processor simpler, but it has some high traffic
loads between the CPU and the memory, and with high traffic comes
lower I/O performance.

Write-back memory, on the other hand, has the CPU updating the
cache during the write, but the actual updating of the main memory is
postponed until the line that was changed in memory is discarded
from cache. At that point, the data that has been changed is written
back to main memory.

If you are looking for a performance comparison, write-through cacheing

provides somewhat better performance because it reduces the number of
writes to main memory. With the performance increase, however, there is a
catch. There is a slight risk that the data may be lost if the system crashes.

Real World Scenario

As you read through this, I am sure you are asking yourself when you may
have to actually do anything with cache. The answer, at least in my experience, is not very often, if at all. This is one of those topics that it is great to
understand, because it will affect how your server performs, but you will
probably never have to "do" anything with it.
Knowing how these things work will help you when you compare specifications on various components. As a rule of thumb, information stored in
memory is faster to access than information stored anywhere else. Therefore, in this case, more is usually better! If you are comparing two RAID
controllers, for example, and one comes with 4 MB of cache, and the other
comes with 16 MB of RAM, you can expect the controller with 16 MB of
RAM to provide better performance. You can also expect it to cost more.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

138

Chapter 4

Memory

Main Memory

Now we have to look at how all the other memory within the system
works.
So, what do you know about main memory so far? Just that the CPU uses
memory to store information and instructions that it may need later. The
more memory you have, obviously, the more instructions or information can
be stored there. But how is it stored there? We have already looked at one
memory technique when we looked at cache. Lets take a look at a couple of
other ways that memory is used. We are going to look at paged memory,
interleaved memory, and shadow memory before we get into error correction, parity, and all that fun stuff. Some of this stuff may not be part of the
objectives, but, for example, you have to understand how paged memory
works before you can understand why interleaved memory is better.

Paged Memory
A year ago, I bought a file server to use in my lab. When I bought the server,
one of the things that the base server was short on was memory. When I
checked the vital statistics on the server, the marketing information said that
it takes just plain old standard memory, so I figured memory is pretty cheap.
I can slap a few DIMMs in there and bring it up to where I want it to be.
Since it is a lab server, it is older and I bought it for a very good price, so I
figured I could add memory without a problem. When I received the server
and also received the technical specifications, it called for fast paged mode
(FPM), error correction code (ECC) memory. That stuff is pricey. Now,
instead of looking at $150 for a 128MB SIMM, I was looking at $500 for a
pair of SIMMs that equal 128 MB. So I did some research and this is what
I found.
Typical memory access by the memory controller is handled in a way that
is similar to reading a book. Just like reading this book, if you want information on paged memory, you access this page. With memory, if it wants
access to certain information, it just accesses the memory page. Once the
page has been accessed, then the information can be gathered in. This process works just great when you are talking about workstations that dont
necessarily have to access information out of memory a lot. When you start
talking about a server, this is a different story. See, with straight paged mode
memory, every time the system wants a bit of information it has to go and
access the appropriate page. With fast page mode memory, this delay is overcome by letting the CPU access multiple pieces of data on the same page,

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Main Memory

139

without having to relocate the page each and every time. This works as long
as the read and the write cycles are on the loaded page.
Fast page mode has certain benefits. For example, there is less power consumption because the pages will not have to be located or sensed each time.
I am pretty sure that the implementation of FPM will not amount to a massive reduction in our electric bill. FPM also has some drawbacks, not the
least of which is price. So, if you can, avoid my mistake, and avoid FPM.
Paged mode memory, on the other hand, simply divides up the RAM in
your system into small addressable groups or pages. The pages can be from
512 bytes to several kilobytes long. The improved memory management on
mainboards has now advanced to the point that it is very similar to fast page
mode, where subsequent memory access from the same page is accomplished
without the CPU having to wait for the memory to catch up. This is referred
to as zero wait state. If the access does take place off the current page, there
may be one or more wait states added while the new page is located.
Now dont confuse paged mode memory with the way Microsoft
Windows 2000 and Novells Netware 5.1 use page files to increase memory.
Before we get into interleaved memory, lets look at that.

Memory Using Page Files

I first came across the concept of page files with Windows NT 4, and I have
to admit I was pretty impressed with the unique solution to an old problem.
The problem with servers has always been that you always seemed to be
short of memory. Back in the early days of local area networks, RAM was
selling for several hundred dollars a megabyte and the mainboards could
only address so much memory. As the CPUs got faster and computers got
smarter, the total amount of usable memory grew, and the price of memory
came down. Of course, at the same time, the operating systems got fatter and
started taking up more memory. So, as a system administrator, you wanted
to provide great response to your users, and you knew that the cheapest way
of doing that was to add more memory, but it didnt take too long before you
simply ran out of room. So there had to be another way. That is where somebody started poking around your average computer and started thinking,
What is cheaper than memory (though not as fast) and can be extended
almost without limits? Disk space! Hey, disk space is a lot cheaper than
DRAM, and it can be in abundant supplysurely there has got to be a way
to use disk space in place of memory. And that was the birth of page files.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

140

Chapter 4

Memory

Page files work like this. When your server comes up, the network operating system (NOS) takes a look at the amount of free space that you have
on your disk subsystem and the NOS then takes part of that free space and
creates what is called a page file. The page file is only for the use of the operating system. This isnt a high-level secret place for network administrators
to store stuff. Now, as we have seen, as the server gets busy, it takes information that it needs, and moves that information into memory. Because it is
a busy server, the length of time instructions can stay in memory may be
exceptionally limited. When it comes time for the information to be flushed
from memory, the system has two choices: It can flush the information from
memory, so the next time it needs that information it can go back to the
application to locate it, or it can move the information to the page file, where
it will be more readily available.

Page files and virtual memory have a language all their own. For example, a
page fault occurs when your system is looking for information in RAM and
cannot find it, so it has to refer to the page file. This is referred to as a page
fault. Page faults then come in two varieties; soft page faults and hard page
faults. Information in memory is stored in frames. When the information is
moved to page files, there has to be a place to temporarily keep that data, and
these are called free frames. The plan is that these frames will be moved into
buffers and then written to disk before replacement data comes along. If a
page fault occurs, and the data is in one of the free frames that has not actually
been written to disk, this is called a soft page fault. If the data has already been
written to disk, it is a hard page fault. Soft page faults are handled more
quickly than hard page faults.

Page files are just near-line storage for information that otherwise would
be stored in memory. So, take a look at Figure 4.4 and you will see how the
CPU uses its memory.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Main Memory

FIGURE 4.4

141

CPU memory access

CPU

Cache

Main memory

Disk

Page files, or disk swapping, are part of a concept called virtual memory.
The virtual memory concept works like this: In a 32-bit computer, the maximum amount of memory that can be conceived is 4GB. The page file system
or the disk swap space is just an area on the hard disk that can be used as an
add-on to the main or physical memory. The page file then is all the memory
that can be used, and the physical memory is the memory that physically
exists. Look at Figure 4.5 and see if that makes it any clearer.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

142

Chapter 4

Memory

FIGURE 4.5

Virtual memory

Virtual memory space

Disk swap
space

Main
memory

Virtual memory space:

All possible memory
addresses
(4GB in 32-bit systems)
All that can be conceived
Disk swap space:
Area on hard disk that can
be used as an extension of
memory
(Typically 100MB)
All that can be used
Main memory:
Physical memory
(Typically 64MB)
All that physically exists

So, with virtual memory space, you are dealing with an address that can
be conceived of but doesnt really correspond to any real memory. If something tries to access it, that attempt generates an error. With page file or swap
file space, if the address is read, the information is on the disk, so it has to be
moved to main memory. This is faster than searching an entire disk because
the memory table has the actual disk location mapped.
Finally, there is the main memory. When the processor wants something
from main memory, it is available immediately.
So, page file memory isnt really memory, kind of. It is virtual memory.
Lets get back to the real stuff and look at interleaved memory.

Interleaved Memory
In the eternal quest to make things faster, the next step up the memory food
chain is interleaved memory. The whole reason for using interleaved memory is
that provides faster response time than paged memory. Check out Figure 4.6.
This is the way paged mode memory accesses information, one step at a time.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Main Memory

FIGURE 4.6

143

Non-interleaved memory

CPU

Bus

Cache

Bus

Memory

Compare that to Figure 4.7, which shows the way interleaved memory
accesses four memory chips.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

144

Chapter 4

Memory

FIGURE 4.7

Interleaved memory

CPU

Bus

Cache

Bus

Memory Bank 3

Bus

Memory Bank 2

Bus

Memory Bank 1

Bus

Memory Bank 0

Interleaved memory combines two banks of memory into one. The first
section of memory is even and the second is odd, so memory contents are
alternated between these two sections. When the CPU begins to access memory, it has two areas that it can go to. With faster processors, they dont have
to wait for one memory read to finish before another one can begin. This
means, for example, that memory access of the odd portion can begin before
memory access to the even portion has completed.
The good news is that interleaving can double your memorys performance. The bad news is that you have to provide twice the amount of memory in matched pairs. Just because your PC says it uses interleaving and
allows you to add memory one bank at a time, do not be confused. The computer is simply disabling interleaving and you may notice a degradation of
system performance.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Main Memory

145

Shadow Memory
Besides the various types of RAM being used on your server, there is also
memory that is read only. Not surprisingly, it is referred to as Read Only
Memory (ROM).

Testing Tip: If you are like me, once you walk into a testing room, you start to
freeze up and question everything, thereby confusing yourself! And, I tend to
forget what it is that certain things dofor example, Read Only Memory. One
of the things that I have found helpful is to pay close attention to what the
words mean, because unlike marketing or management speak, computerese
tends to be very descriptive. I mean, when you see ROM, if you know the acronym stands for Read Only Memory, you have a really good clue what that stuff
is used for. If it had been named by someone in marketing or management, it
would have been called something like silicon-enhanced, integrated longterm memory paradigm used only for perusal and not for continuous reconfiguration in this regard unless we have at least three meetings. You get my
point!

ROM devices are things like the Basic Input/Output System (BIOS) on
your mainboard. These devices tend to be very slow, with access times in the
several hundreds of nanoseconds. Because your CPU is much faster than
that, ROM access requires your CPU to go through a large number of wait
states before returning instructions, and that just slows down the whole systems performance. How big of a deal is that? Well, think of the things that
have their own BIOS:


Mainboards

Video cards

SCSI controllers

These are things that will be accessed very frequently, so you can see where
it could become an issue. Some computers use a memory management technique called shadowing. When shadowing is employed, the contents of
ROM are loaded in to an area of the faster RAM during system initialization. Then the computer maps the fast RAM into memory locations used by
the ROM devices. After that is done, whenever the ROM routines have to be
accessed, the information is taken from the shadowed ROM rather than
accessing the actual IC. In this way, the performance of the ROM can be
increased by more than 300%.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

146

Chapter 4

Memory

Real World Scenario

When you start talking about main memory, there are some tips that you
should keep in mind, especially when dealing with a server.
As I have alluded to time and again in this book, the two key factors of a
server are availability and performance. One of the ways that you can
impact both these factors is with judicious use of RAM.
Most server platforms will be designed to handle a specific type of memory.
When you buy your server, load up on as much RAM as you can afford from
the manufacturer. When the server arrives (or when you document your
server) make sure to document what type of memory, manufacturer, part number is installed, as well as the size and speed of the SIMM. You should also document how many (if any) open slots you have available for expansion.
If you notice that performance on your server has degraded, do a baseline
check on how memory is being used. This can be accomplished in many
ways, depending on the operating system. If your NOS uses a page file,
check to see how much the page file is being accessed. If it is getting
accessed very, very often, you need to add memory to your server. Making
sure that you add the right kind of memory to the server will insure that the
server will operate at peak efficiency and provide you with the most bang
for your buck!

Memory Error Checking

If you pause a minute and take a look at the big picture, you are going
to see that we are talking about some pretty serious stuff. We are talking
about information sets on how the computer will operate, as well as program
information, and data is being moved into and out of memory at a rapid rate.
If that information is not moved correctly, nothing works properly, and your
life is not very much fun. So, it is vitally important that all of the instructions
and all of the data remain error-free. Think about all the things that can
result in corrupt instructions:


Electrical noise

Component failure

Corrupt drive information

Video problems

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Memory Error Checking

147

Parity in the memory subsystem works like this: When a byte is written to
memory, it is checked, and a ninth bit is added to the byte as a checking or
parity bit. When the CPU needs to access the information from memory, the
CPU runs the numbers and calculates the expected parity bit. At that point,
the parity bits are compared and, if they match, the information is deemed
correct. If the parity bits do not match, the system comes up with an error
and, depending on the sophistication of the system, it may actually halt.
Every byte is given a parity bit. If you are working with a 32-bit PC, there are
4 parity bits for every address. If the PC is a 64-bit model, the number of parity bits increases to 8.
There are two types of parity: even parity and odd parity. With even parity, the parity bit is set to 0 when the number of 1s in the byte is even. That
will keep the number of 1s in the calculation even. If the number of 1s in the
byte is not even, then the parity bit will be set to 1, thus making the number
of 1s even.
The reverse is true with odd parity. In this case, the system wants to make
sure there is always an odd number of 1s in the byte. So, if the number of 1s
in the byte is odd, the parity bit is set to 0. If the number of 1s in the byte
is even, the parity bit restores order by being a 1.
If you look at this, you are going to notice that even and odd parity are
exactly opposite, and that is OK. It does not matter in the greater scheme of
things.
Like most things that are simple and are free, parity has some shortcomings.
First of all, when it discovers a problem, it cannot fix the problem. It only
knows that one of the bits in the byte has changed, but it doesnt know which
bit and it doesnt know if it changed from 0 to 1 or from a 1 to a 0. Also, what
happens if 2 bits are corrupted? If a 0 gets changed to 1 and another 0 gets
changed to a 1, as far as parity is concerned everything is wonderful.
Given this scenario, like most things in computing, someone decided there had
to be a better way, and that better way was called Error Correction Code (ECC).

ECC Memory
Like everything else, memory schemes evolve, and people whose priorities
are high availability and high reliability understand that higher cost usually
follows. ECC memory works in conjunction with the mainboard memory
controller to add a number of ECC bits to the data bits. Now, when data is
read back from memory, the ECC memory controller can check the ECC
data read back as well.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

148

Chapter 4

Memory

This means that ECC memory is superior over memory with just parity
for two reasons. First, ECC memory can actually correct single-bit errors
without bringing the system to a halt. It can also detect when there have been
2-bit, 3-bit, or even 4-bit errors, which makes it a very powerful detection
tool. If there is a multi-bit error detected, the ECC memory will report the
error and the system will be halted.
There is some additional overhead with ECC. It takes an additional 7 or
8 bits to implement ECC.

Have you ever wondered if there was a way to determine whether your system has parity or ECC memory? There is. All you have to do is count the number of memory chips on each module. Parity and ECC memory modules have
a chip count that is divisible by 3. Any chip count not divisible by 3 indicates
that the memory module is non-parity.

Extended Data Out (EDO) Memory

Several times during the first four chapters I have mentioned that if you
really want to know what something does, take a good close look at the
name. This is another example. Extended Data Out (EDO) RAM just keeps
information hanging around by lengthening the time the information is valid
in memory; therefore the datas existence on the data bus is extended. This
is done by changing the DRAMs output buffer and prolonging the time that
the read data is valid. With EDO, the read data is valid until the motherboard signals that it doesnt need the information any more. This helps ease
some time constraints on the memory and also allows an improvement in
performance with little or no increase in overhead or cost. There is an external signal that is needed between the motherboard and the memory, so the
motherboard must have an EDO-compliant chipset.
EDO RAM can be used in non-EDO motherboards, but obviously there
will be no performance improvement.

Unbuffered vs Buffered vs Registered

As we continue with everything you ever wanted to know about memory but
didnt know who to ask, we are now going to look at how the memory controller accesses and writes information onto the memory chip. This is done
in one of three ways: unbuffered, buffered, or registered.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Memory Error Checking

149

Unbuffered memory
Unbuffered memory talks directly to the chipset controller. There is nothing
standing between the memory module and the controller. Therefore, information is written quickly to memory, with very little overhead.

Buffered
Buffered memory is a DIMM that has a buffer chip on it. If you are using a
DIMM with lots of chips on it, it requires a lot of effort on the part of the system
to write information into memory. Some manufacturers will use a re-drive
buffer on the DIMM to just boost the signal and reduce the load on system. The
buffers are overhead and therefore they introduce a small delay in the electrical
signal.

Registered
With registered memory, the DIMM contains registers that will re-drive or
enhance the signal as it goes through the memory chip. Because the signal is
being enhanced, there can be a greater number of memory chips on the
DIMM. Registered memory and unbuffered memory cannot be mixed.
Just like buffered memory, registers slow things down. Registers delay
things for one clock cycle to make sure that all communications from the
chipset have been collected. This makes for a controlled delay on heavily
used memory.

Hardware Compatibility Lists

Whenever I think of hardware compatibility lists (HCL), I think of one of my
favorite presents, which came from my youngest daughter. I cant remember
if it was for birthday, Fathers Day, or just to shut me up. You see, my youngest daughter is a tech-weenie wannabe. She is not there yet, but she works
with her mother, who is totally computationally challenged, in an office
where the entire technological skill level is understanding how to turn the
computers on. Denise has been forced into the role of on-site technical person while I am off traipsing around the country teaching people how to build
and administer networks. Anyway, as Denise grew in her role, she kept calling with questions. Since I was usually on a short break from the classroom,
I usually didnt have much time to spend with her, so we would do the quick
fix and if that didnt work, my next response would be, So, what did the
manual say? She began to tire of that, and like most people in their 20s
when dealing with their parents, she was not afraid to let her exasperation
show. So, we simplified it. She would ask a question and I would reply,
RTFM, meaning Read The Fine (or some such) Manual. This was not an
Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

150

Chapter 4

Memory

answer she liked, but she did it, and lo and behold, the number of calls
decreased and her skill level increased. She began to see the humor in the
whole thing, and that is why she gave me a shirt monogrammed with RTFM,
and the note that said I should wear that during difficult classes.
Now this really is a do-as-I-say-and-not-as-I-do situation, because I have
been there, done that, got the T-shirt, and therefore should not have to read
the manual. Every time I take that attitude, I am immediately shot down by
doing something incredibly stupid (and usually costly) to prove the point.
So, let me put it to you this way. Whether you have just gotten a new copy
of the SuperWhizBang 6000 Operating System, or you need to put a card in
a computer, it never hurts to check the hardware compatibility list to see if
that card will actually work in the system. Or, if you really want to be daring,
you can read the compatibility list before buying the card, thus saving yourself time and frustration. These things are written for a reason, and they are
usually on the Internet or come with the program. Check to make sure your
system meets minimum requirements and you will save yourself tons of
headaches later.

Real World Scenario

Here is another case of true confessions. Earlier in the chapter, I talked about the
lab server that I have that uses really expensive memory. Well, truth be told, I
have a really well equipped lab, including several components that have not
been installed into servers and will probably never be installed in servers. My
wife keeps referring to that as wasted money. I tend to think of it as a lesson that
I should have learned, and probably didnt
Hardware compatibility lists are wonderful things. You see, if I would take
the time to check the HCL for a particular NOS, I would have found out that
many of my orphaned components would not work with the NOS. If I had
checked the HCL before I took advantage of the great deal that I found, I
would have saved myself tons of money. Unfortunately, I tend to purchase
first and check later, which is really a bad habit.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

New Stuff

151

Now, when I am buying for my lab, the worst thing that is going to happen is that
I will have a stack of components on my bakers rack that will probably never see
the inside of a server. If I were doing this at a client site, I would have wasted the
clients money and time, not to mention shooting my credibility! When I make a
purchase for a client, I am very careful to make sure that the component appears
on the HCL. If the proposed solution does not appear on the HCL, I then check the
manufacturers web site to see if there is support available. If there is alleged support available, I then download the drivers and test the component in a lab
machine before trying to install it in a production environment. Remember, we
want our servers to be high availability, and if we take a server down, we have to
make sure that the time out of service is used to the best advantage.

New Stuff

here are some other memory technologies whose names you may run
into that we havent covered here: Rambus (RDRAM) memory, Double Data
Rate SDRAM (DDR SDRAM), and IBM Memory Expansion Technology
(MXT). Since two of these types of memory require special mainboards, it is
important that you know what the specifications mean before you fill out a
purchase order for new memory for your server.

RDRAM
RDRAM is an Intel invention that got off to a rocky start. And its life hasnt
been too great either. Rambus was originally supposed to be the next great
memory advance, but then it got bogged down in life. Delivery was late,
there were squabbles between Intel and memory manufacturers that led to
lawsuits, and then when Rambus finally did hit the market, performance was
nowhere near expectations. In published reports, Intels own benchmarks
showed that less-expensive SDRAM technology running at 133MHz outperformed RDRAM running at 800MHz.
When Intel brought RDRAM to market, they wanted the manufacturers
to pay a licensing fee to Intel for the technology. Well, since the margin in
memory is nonexistent and the traditional memory shopper is looking for
price as well as performance, this strategy did not go over well.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

152

Chapter 4

Memory

DDR SDRAM
DDR SDRAM is like normal SDRAM in many ways. For example, it works
with the front-side bus clock in the system. The memory and the bus run
instructions simultaneously. This means that, as bus speeds have increased,
so has system performance.
The big difference between the two is the way that DDR reads the data.
It has found a way to effectively double the speed of the SDRAM. This means
that if the data rate is usually 133MHz, DDR will transfer data at a rate of
266MHz.
DDRs also come in DIMMs, but they will not fit in the standard SDRAM
slot so you have to use a specially designed mainboard. The same problem
with configuration carries over to the laptop market. The SO DIMMs will
need a specially designed mainboard also. The DIMMs will have different
notchings and a different number of pins.
DDRs come in ECC for servers, and non-ECC for workstations.

DDR vs Rambus
In a study done by InQuest Market Research in November 1999 (http://
www.inqst.com/ddrvrmbs.htm), it was reported that the performance differences were negligible between Rambus and SDRAM.
Yeah, but that is SDRAM. What are the performance statistics for Rambus and DDR? InQuest used a benchmark called the SteamD that has been
released by the University of Virginia. This benchmark is designed to evaluate the bandwidth of memory to the processor. The margin of error for this
benchmark is less than 1%. In this study, DDR beat out Rambus by a significant margin in all tests, exceeding 30% in some cases and averaging 24.4
% performance advantage for this benchmark.
There is another version of the testing suite, this one to show the memory
types that work with Windows. This benchmark is WSTREAM.EXE.
According to the developers, the compound precision error rate is in the
range of 30%, and the developer has said that the program is inaccurate
under Windows NT 4. In tests using Windows 98, InQuest showed that the
DDR performance advantage had decreased to just 2.7%.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Increase Memory

153

Keep in mind, this study was done in the fall of 1999, and the way memory
technology has changed, all bets could be off by the time you read this. As a
matter of fact, at the time this study came out, neither Rambus or DDR had
been released to the general public yet. Do your research before filling out the
purchase order for any new technology.

IBM Memory Expansion Technology (MXT)

In June 2000 IBM introduced an MXT, which IBM claimed could effectively
double the amount of memory in servers based on Intel processors. MXT is
based on a new cache memory design that compresses data down to half its
size. IBM is predicting that this will save large implementations thousands
and sometimes hundreds of thousands of dollars. In IBMs example, using
MXT with a typical rack-mounted server configuration using Windows
2000 can allow an implementation with 84GB of memory to act like a system with 168GB of memory, saving the company about $250,000.

More information on IBM research is at www.research.ibm.com.

Increase Memory

et me make a very bold statement. Id say that at least 2550% percent of

todays servers are RAM-starved. What I mean by that is that they dont have
enough system memory to handle the load theyre under. Oftentimes, when a
server is RAM-starved, it can use physical disks as an extra memory source. This
technique is called paging or virtual paging. But when a server is paging, the
speed of the system slows dramatically down. It is always better to have a server
utilize system memory than to have it page to disk for part of its memory.
Usually its easy to cure RAM-starvationsimply add more system RAM.
But there are some concerns youll have to address before going forward.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

154

Chapter 4

Memory

Verifying Hardware and OS Support

The first and probably most important question youll have to get the answer
to is whether the current computer has any space available for RAM
upgrades. Sometimes a servers RAM slots are maxed out and you cant add
any more RAM. In an instance like that, your next option is to replace the
server or see if you can replace some slower peripherals with newer faster
models. While peering into the computer to find out if youve got available
open slots, check to see what kind of memory is currently installed in the
computer and how many open slots you have.
Youll also want to check the specifications of your operating system to
make sure it can work with the amount of RAM youre planning on adding.
Generally this isnt going to be an issue, but its worth checking anyway.

Verify That Memory Is on Compatibility Lists

After determining that you have open slots and making a decision about
how much memory youre going to order and from which manufacturer, you
should check the compatibility list for your server to make sure your memory
decisions are appropriate. Some computer manufacturers require that you
use their memory in their systems. If you try to buy memory that they havent
officially blessed, you may run into trouble. Its very important, especially
with tier 1 vendors (such as Dell, IBM, Compaq, HP, and others) that the
memory youre planning on purchasing will work in their computer.
I dont think its wise to trust third-party clone memorymemory thats
designed to emulate proprietary memory chips that manufacturers recommend in their systems. Ive had very mixed luck with clone memory chips and
tend to prefer to use the real McCoy even though its a bit more expensive.
Why the care to make sure the memorys compatible? Because if you
have memory troubles on down the road, you might have a lot of trouble
diagnosing the problem. Its important to always be sure that youre putting equipment into the computer that the manufacturer has officially
said will work. Otherwise, youre setting yourself up for many problems.

Verifying Memory Compatibility

The next item on the list has to do with the kind of memory youre going
to buy. Its not enough to know that you want to buy a certain kind of memory
with certain specificationsyoull also need to know what youre pairing the
new memory with in the computer. For example, if your system currently has
72-pin 100 nanosecond (ns) DIMMs (dual inline memory modules) inside it,

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Increase Memory

155

you cant go putting 70ns DIMMs in the computer and hope that things work
correctly. As a general rule of thumb, youll want to closely match whats
already in the computer. There are several things to consider:
Capacity What is the capacity of the RAM thats currently in the computer? What is the maximum RAM capacity that the computer is capable
of handling? If your computer can handle a maximum of 128MB of RAM
and youve already got 64MB in the computer you can only add 64 more
megabytes to the computer before its satiated.
Brand If your manufacturer documentation doesnt have any particular
brand in mind for RAM upgrades, be sure that you pick a known reputable
vendor for your RAM. Dont try to short sheet your server by purchasing
from an unknown vendor so you can save a buck or two. Youll likely find
that the RAM doesnt work correctly and that youll have lots of problems
with it.
Speed What is the speed of the RAM, in nanoseconds, thats currently in
the computer? You cannot mix and match RAM speeds. Its vital that you
match the RAM speed currently in the computer with the speed youre
planning on adding.
EDO Extended Data Output (EDO) RAM has the capability of retrieving
the next block of data at the same time as its sending the previous data
block to the CPU. Do not mix and match EDO and non-EDO RAM. You
might experience difficult-to-diagnose erratic activity with the computer
after upgrade.
ECC/Non-ECC Error Correcting Code (ECC) memory has the ability
to check the validity of the data as its passing into and out of the chip. Its
not as vital to make sure you dont mix up ECC with non-ECC memory.
You may want to consider purchasing all ECC memory for your server
and throwing away any non-ECC chips you might encounter.
SDRAM/RDRAM Synchronous Dynamic RAM (SDRAM) has the
capability of running substantially higher clock speeds than older RAM
chips. Newer SDRAM chips can run at a systems 100MHz bus speed,
thus producing significantly faster throughput. But they bog down when
running much faster than 100MHz. Rambus Dynamic RAM
(RDRAM), a RAM chip invented by Rambus, Inc. (www.rambus.com)
can run at phenomenally higher clock speedsa maximum of 600MHz
as of this writing. Thus, as newer system buses come out that are capable
of running at higher clock speeds, RDRAM can keep up with the activity. Another kind of RAM, a competitor to RDRAM being designed by

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

156

Chapter 4

Memory

a consortium of computer manufacturers and called SyncLink Dynamic

RAM (SLDRAM), will be introduced soon.
Contacts The kinds of contacts used on the RAM chips are important as
well. For example, some RAM chips have gold contacts. It isnt vital to
match the type of contact currently in the system, but understand that gold
provides a better electronic contact point than inferior conductor materials
such as aluminum.
You must ascertain the kind of RAM currently in the system and compare
with the kind youd like to purchase for the computer to make sure that you
wont run into any conflicts between the old and the new.
You should also note whether you need to buy the RAM in pairs or not. In
some computers the RAM is installed in pairs of chips while in others you can
buy a single chip. For example, some servers allow you to purchase one DIMM
at a time while other computers require that you purchase two DIMMs (and
occasionally may require that you install them in matching slots).

Oftentimes a computer manufacturers Web site will list the kind of memory
that originally shipped with the computer, thus giving you some documentation that you can utilize when purchasing compatible additions.

Perform an Upgrade Checklist

As always, its a great idea to write out a checklist that you can use. In
your checklist, put a place for the old RAM so that you can write down all
of its characteristics. Then supply a place where you can match the new
RAMs characteristics with the old.
If in doubt about a RAM chip, call the vendor or manufacturer for information on the chip. If you cant ascertain a chips characteristics, Id advise
buying all new chips and forgetting about pairing with the old ones.

Verify That Server and OS Recognize the Added Memory

Servers (computers in general) are really funny about adding memory. You
put the new memory in the computer and watch carefully as you start it up.
Youll generally see the computers power-on routines count the memory
and note the changes. But then you might be surprised to see an error on the
screen telling you that the BIOS has encountered memory mismatch errors.
Youll be prompted with a key to hit to enter the BIOS configuration utility.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Summary

157

This isnt any big deal. Just go into the BIOS, verify that the new memory
size has registered, and then exit, saving changes (being careful not to change
any other BIOS options!). The server will restart and this time youll see it
successfully count and pass through power-on without generating any more
errors.
Once the OS has loaded, verify that it sees the correct amount of memory
as well. If you encounter any problems, note any errors that are reports in the
logs. Ive never had a problem with an OS not recognizing the proper
amount of RAM if the BIOS has successfully noticed and registered it.

Perform Server Optimization to Make Use

of Additional Ram
Finally, there are some things you can do OS-wise to make sure that the additional RAM is utilized properly. One thing that you can check is whether applications are optimized to use the correct amount of system memory. Oftentimes
applications will have configuration settings that allow you to tune how much
physical memory can be used at any one time. Once you upgrade memory, it
may be time to check these settings and make sure theyre still correct, or update
as needed.
Also you can check the page file size and cut it back a notch or two. You
may also be able to tune the virtual paging settings so that the system makes
optimal use of RAM rather than paged memory.

Summary

his chapter has centered on memoryall different forms of it and all

different uses for it. The information you have just gone over should make
it much easier for you to determine the proper type of memory to put into
your server to get peak performance. Memory is one of the key ingredients
to all server operating systems. In most cases, when you want a performance
increase, whether at the CPU or in main memory, the slogan seems to be,
Add more memory.
Before increasing the memory in a server, its important to ascertain the kind
of memory thats currently in the computerhow much capacity, the speed of
the RAM, what kind of RAM it is. You should also validate with the server manufacturer what brands of RAM are supported for addition to the server. Some
manufacturers prefer that you use their brand of RAM and look with disfavor

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

158

Chapter 4

Memory

on third-party RAM manufacturers. Check your documentation! There are key

things you need to match when shopping for new memory: the memorys speed
and whether its EDO, ECC, SDRAM, etc.
You also need to verify that the BIOS is aware of the addition and checking
to make sure the OS sees the new RAM upgrade was well. We also discussed
optimizing RAM usage by tweaking applications to use more RAM and by
adjusting virtual paging settings.

Exam Essentials
Know the differences between L1 and L2 processor cache. L1 cache is
actually on the processor. L2 cache is usually part of the mainboard and
is used exclusively for the processor, but it is not part of the processor. L1
and L2 cache are measured in kilobytes.
Know why Raid uses cache. RAID systems use cache to improve
throughput and speed up disk reads and writes. RAID cache is measured
in megabytes.
Know the difference between write-back memory and write-through
memory. Write-through memory writes information to cache and to
main memory at the same time. Write-back memory has the CPU updating the cache during the write, but the actual updating of the main memory is postponed until the line that was changed in memory is discarded
from cache.
Know how memory interleaving works; know how paged memory
works. When your server comes up, the network operating system
(NOS) takes a look at the amount of free space that you have on your disk
subsystem and the NOS then takes part of that free space and creates what
is called a page file. The page file is only for the use of the operating system. The page file is then used to hold information from memory that may
be used again in the near future.
Know the difference between page faults, soft page faults and hard page
faults. A page fault occurs when your system is looking for information
in RAM and cannot find it, so it has to refer to the page file. This is
referred to as a page fault. Page faults then come in two varieties; soft page
faults and hard page faults. Information in memory is stored in frames.
When the information is moved to page files, there has to be a place to
temporarily keep that data, and these are called free frames. The plan is
that these frames will be moved into buffers and then written to disk
before replacement data comes along. If a page fault occurs, and the data
is in one of the free frames that has not actually been written to disk, this
Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Summary

159

is called a soft page fault. If the data has already been written to disk, it
is a hard page fault. Soft page faults are handled more quickly than hard
page faults.
Know the difference between ECC memory and EDO memory. ECC is
error-correcting memory. EDO memory just lengthens the amount of
time information can be stored in memory before it is sent to a page file,
or discarded.
Know the difference between unbuffered, buffered and registered memory.
Unbuffered memory writes information directly to the chipset controller.
Buffered memory uses buffered chip to boost the signal and ease the strain
on the system. With registered memory the DIMM contains registers that
will re-drive or enhance the signal as it goes through the memory chip.
Know when to use a hardware compatibility list. Whenever you add
hardware to a server, check the NOS hardware compatibility list. If the
component does not appear on the HCL, check the components manufacturers web site to make sure the appropriate drivers are available.
When in doubt, dont install the device.
RAM upgrade. Know and understand how to upgrade system RAM
and what components to check for when shopping for upgrade RAM.

Key Terms
Before you take the exam, be certain you are familiar with the following terms:
Basic Input/Output System (BIOS)
buffered memory
cache memory
Double Data Rate SDRAM (DDR SDRAM)
Dual Inline Memory Module (DIMM)
dual inline package (DIP)
Dynamic Random Access Memory (DRAM)
error correction code (ECC)
even parity

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

160

Chapter 4

Memory

Extended Data Out (EDO)

fast paged mode (FPM)
hard page fault
hardware compatibility list (HCL)
interleaved memory
kilobyte
Level 1 cache (L1)
Level 2 cache (L2)
main memory
megabyte (MB)
Memory Expansion Technology (MXT)
non-ECC
odd parity
page fault
paged mode memory
parity
RAID cache
Rambus Dynamic RAM (RDRAM)
Random Access Memory (RAM)
Read Only Memory (ROM)
registered memory
Single Inline Memory Module (SIMM)
Small Outline DIMM
soft page fault
Static Random Access Memory (SRAM)
Synchronous Dynamic RAM (SDRAM)
SyncLink Dynamic RAM (SLDRAM)

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Summary

unbuffered memory
write-back memory
write-through memory
zero wait state

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

161

162

Chapter 4

Memory

Review Questions
1. What does DIP stand for?
A. Dual internal processors
B. Dynamic Induction Processing
C. Dual Inline Package
D. Dynamic Inline Package
2. SIMMs came in which pin configurations?
A. 28-pin
B. 30-pin
C. 64-pin
D. 72-pin
3. With how many data bits were computers working when the 30-pin

SIMMs first came out?

A. 16
B. 32
C. 48
D. 64
4. What was the minimum number of 30-pin SIMMs that you needed to

install to account for the proper number of data bits?

A. 1
B. 2
C. 4
D. 8

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

163

5. How many data bits were supported on each 72-pin SIMM?

A. 32
B. 36
C. 72
D. 144
6. What was the minimum number of 72-pin SIMMs that needed to be

installed to account for the proper number of data bits?

A. 1
B. 2
C. 4
D. 8
7. Thirty-pin SIMMs were installed in memory banks. What were the

names of these banks?

A. Bank 1 and Bank 2
B. Bank 0 and Bank 1
C. Bank A and Bank B
D. First Bank and Second Bank
E. U.S. Bank and Wells Fargo
8. What is DIMM an acronym for?
A. Dips used In Memory Modules
B. Dual In-place Memory Maker
C. Dual Inline Memory Maker
D. Dual Inline Memory Module

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

164

Chapter 4

Memory

9. Other than physical appearance, how is a DIMM different than a SIMM?

A. Each DIMM supports a minimum of 64MB of memory.
B. Each DIMM has 32 DIPs on it.
C. The opposing pins remain separate and isolated to form two contacts.
D. DIMMs use only gold contacts where SIMMs use either silver or gold.
E. DIMMs hold cache memory; SIMMs hold RAM
10. What type of computer uses DIMMs?
A. 8-bit
B. 16-bit
C. 32-bit
D. 64-bit
11. What type of DIMM is used in a laptop?
A. Double-edged DIMM
B. 64MB
C. SO DIMM
D. SEW DIMM
12. How many pins are on the DIMM used in servers?
A. 168 pins
B. 164 pins
C. 162 pins
D. 160 pins
13. There are two types of RAM. Please select them from the list below.
A. DIMM
B. SIMM
C. DRAM
D. SRAM

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

165

14. Bonnie is attempting to add some system memory to a server. The

systems bus accepts 72-pin DIMMs. Bonnie is attempting to add

one new DIMM to the existing system. Once she adds the new
memory she gets no errors, but the computer doesnt recognize the
new memory. What could be causing the problem?
A. Incorrect speed on new DIMM.
B. Brand of DIMM isnt compatible with rest of architecture.
C. Incorrect capacity of new DIMM.
D. Current DIMMs arent ECC.
E. Old DIMMs have silver contacts.
15. Suzanne is working on a server that has four slots in it for DIMMS.

Two of the slots have 64MB DIMMS in them already. Suzanne wants
to add a 128MB DIMM, giving the system a total of 256MB of total
system memory. When she adds the DIMMS, the power-on self-test
memory count shows the full 256MB but she now gets an error telling
her to adjust the BIOS. What could be the problem?
A. Nothings wrong.
B. Cant pair DIMMS of different capacities.
C. First two DIMMS are ECC DIMMS, new ones not.
D. First two DIMMS are silver-tipped, new ones not.
16. You are going to install interleaved memory. What is the minimum

number of memory modules you must install to make sure that

interleaving is actually working?
A. One
B. Two
C. Four
D. Eight

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

166

Chapter 4

Memory

17. What does ECC stand for?

A. Error Correcting Cache
B. Error Checking Cache
C. Error Checking and Correcting
D. Error Correcting Code
18. How does ECC differ from parity checking?
A. Parity checking can correct a single bit that has been changed.

ECC cannot.
B. ECC is cheaper because the code is actually embedded into a code

chip on the memory module.

C. ECC memory can actually correct single-bit errors. It can also

detect when there have been 2-bit, 3-bit, or even 4-bit errors.
D. There is no difference.
19. You have a memory module with nine chips on it. What kind of

memory is it?
A. Either parity or non-parity
B. Non-parity only
C. ECC only
D. ECC or parity
20. You have a server that is RAM-starved. You purchase a DIMM from

a reputable memory manufacturing company, install it, and find that

the system wont boot up. What could be the problem?
A. The type of memory you bought isnt supported by the computer

manufacturer.
B. System requires DIMMs to be installed in pairs.
C. Youve exceeded the systems memory capacity with the DIMM

youre adding.
D. System BIOS needs to be adjusted.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Answers to Review Questions

167

Answers to Review Questions

1. C. A DIP is a dual inline package.
2. B, D. SIMMs came in 30-pin and 72-pin configurations.
3. B. When the 30-pin SIMMs first came out, computers were working

with 32 data bits.

4. C. Each SIMM handled 8 data bits, so you had to install SIMM in

groups of 4.
5. A. Each 72-pin SIMM supported 32 data bits.
6. A. With 72-pin SIMMs, just one module would provide the necessary

32 data bits.
7. B. The two banks were referred to as Bank 0 and Bank 1.
8. D. DIMM stands for Dual Inline Memory Module.
9. C. With a DIMM, the opposing pins remain separate and isolated to

form two contacts.

10. D. DIMMs are used in 64-bit computer configurations. This relates to

the Intel Pentium or the IBM RISC processor.

11. C. Small Outline DIMM or the SO DIMM. This DIMM is like a 72-

pin SIMM in a reduced size. It is designed primarily for laptop computers.

12. A. There are 168 pins on a DIMM.
13. C, D. The two types of RAM are Dynamic Random Access Memory

(DRAM) and Static Random Access Memory (SRAM). Choices A and

B are types of memory modules, not RAM.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

168

Chapter 4

Memory

14. A, B, C, D. If Bonnie is trying to put a 100ns DIMM in a system that

has 70ns DIMMs currently installed, she could run into trouble.
Also, if the server shes trying to upgrade has proprietary memory in
it, she could create some problems by not buying manufacturerrecommended DIMMs for the system. Additionally, its not a wise
idea to match ECC with non-ECC memory and so forth. Generally
its a good idea to ascertain whats currently in the system and match
accordingly. The kind of contacts each DIMM has shouldnt affect
the systems operation.
15. A. In almost all cases, after you add memory to a system, you have to

go into the system BIOS and acknowledge that the current memory
count is correct.
16. B. Interleaved memory combines two banks of memory into one, and

therefore you must have at least two modules.

17. D. ECC is the acronym for Error Correction Code.
18. C. ECC memory can actually correct single-bit errors without bringing

the system to a halt. It can also detect when there have been 2-bit, 3-bit,
or even 4-bit errors, which makes it a very powerful detection tool.
19. D. Where the number of chips is divisible by 3, it can be either ECC or

parity memory.
20. A, B. First of all, you should always consult the manufacturers guide-

lines before upgrading RAM in a system. Some RAM that can be

bought from third-party vendors simply wont work very well in some
systems. Some systems require that you install memory in pairs. Youll
have a RAM bay A and RAM bay B and youll have to buy two
DIMMs and place one in A and one in B to make it work. Either of
these situations, or both, could be your problem.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Chapter

System Bus
Architecture
SERVER+ EXAM OBJECTIVES COVERED IN
THIS CHAPTER:

3.7 Upgrade peripheral devices, internal and external


Verify appropriate system resources (e.g., expansion slots,

IRQ, DMA, etc.)

Perform upgrade checklist including: locate and obtain

latest test drivers, OS updates, software, etc.; review FAQs,
instructions, facts and issues; test and pilot; schedule
downtime; implement using ESD best practices; confirm
that the upgrade has been recognized; review and baseline;
document the upgrade.

3.9 Upgrade service tools (e.g., diagnostic tools, EISA

configuration, diagnostic partition, SSU, etc.).


Perform upgrade checklist including: locate and obtain latest

test drivers, OS updates, software, etc.; review FAQs,
instructions, facts and issues; test and pilot; schedule
downtime; implement using ESD best practices; confirm that
the upgrade has been recognized; review and baseline;
document the upgrade.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

h, yeah, now we are a getting into it. If you take the case
off a server, one of the first things you are going to notice is that big green
board that everything else plugs into. Call it a mainboard, call it a motherboard, call it whatever you want, that is where all the information must
pass to go anywhere. Everything else we have been talking about wont
work at all if there is something wrong with the motherboard. In this chapter we are going to look at what sets this component apart and how all the
various parts come together to communicate.

For complete coverage of objective 3.9, please also see Chapter 9.

System Bus Architecture

If you have taken the A+ exam, some of the information we are going to
cover will probably be review, but that is not a bad thing. Review can usually
help us all! We are going to start this section by talking about bus basics and
then move into the way the Peripheral Component Interconnect (PCI) local
bus works. At that point, we will cover most of those objectives listed above.
All of these topics relate to information flow and the speed with which it
moves through the server and out to the network.

Bus Basics
You may have noticed that in Chapter 4, Memory, there were several references to the bus speed that went without explanation. My reasoning was
that I would use that as a promo for the good stuff in Chapter 5. So, what is
a bus? A bus is a set of signal pathways that allow information to travel
between the components that make up your computer. These components can

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

System Bus Architecture

171

be inside the computer or, as we saw in the discussion of SCSI in Chapter 1,

even outside the computer. When we talk about buses, we are generally talking
about one of three types: the external bus, the address bus, and the data bus.
In Chapter 4 we talked about cache memory and the fact that L2 cache
was external to the processor. The processor will use an external bus to get
information to and from the cache memory. The external bus is what the
CPU uses to communicate with all the other peripherals in the computer.
When the CPU wants to talk to a device, it will make use of the address bus.
In this case, the CPU selects a particular memory address that the device is
using, and sends information to that address using the address bus. When the
device responds to the CPU, it will use the data bus. How does the device get
the CPUs attention? Just like a small child, it uses the theory of interrupts.

This discussion is going to make use of some terms that were defined and discussed in earlier chapters: DMA and Bus Mastering. Because these are important
concepts, we will take a couple of sentences to review them here. A Direct Memory Access channel is a channel that a peripheral device can use to write specifically to a set memory address. DMA channels cannot be shared, and the device
does not use the CPU to access memory. Bus Mastering is similar. It is the ability
of a device to perform its function without needing to access the CPU. It writes
information to memory without accessing the CPU.

Interrupts
Interrupts are amazing things. When you have installed a component properly,
it is up to the interrupt to get the attention of the CPU when the component has
information or data to send. If the card is not installed properly, and you have
chosen an interrupt that is already being used, the card will either not function
or the system will completely lock up. Fortunately, with the PCI Bus, each
expansion slot (rather than the card) is assigned an interrupt, so the problem
of misconfigured components has been minimized.
When a card or peripheral has some data to send, it uses something called
an interrupt requestor (IRQ) line. The IRQ is kind of like a student in class
holding up her hand to get the attention of the instructor. In this case,
though, the peripheral is trying to get the attention of the CPU.
Each type of bus has several different types of IRQs, and some of these
IRQs are reserved. For example, IRQ 0 and IRQ 1 are used by the processor
for special processor stuff. The other IRQs can be allocated depending on the
peripherals that are installed.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

172

Chapter 5

System Bus Architecture

IRQs are finite, meaning there are only a few that can be used. If your server
has several different peripherals that are not PCI, you could conceivably run
out of IRQs.

Interrupts are part of the expansion of a PC. Adding a card or a component

into a connector (called an external bus) on the motherboard usually does
expansion. The external bus that you are most likely familiar with is the expansion bus. The expansion bus is that motherboard component that will allow you
to expand the services provided by the computer simply by plugging in a specially designed circuit board. These boards are plugged into connectors or
expansion slots, and this connection allows them to communicate with the CPU.
The expansion buses come in various flavors; for example, there were the
old 8-bit and 16-bit ISA Buses, and there is the PCI Bus, the EISA Bus, and
the Micro Channel Architecture from IBM. In this chapter we are going to
look at PCI and EISA primarily, but it never hurts to have some background.

Expansion Slots
If you were to look closely at an expansion slot, you would see that it is made up
of several tiny copper finger slots. Each finger slot has a row of very small channels that make contact with the fingers on the expansion circuit board. These
finger slots are then connected to the pathways on the motherboard, and each of
the pathways has a specific function. One of the pathways provides the power
necessary to run the expansion card. Another set of pathways is the data bus,
which, as the name implies, transmits data to and from the processor. Another
set of pathways makes up the address bus. The address bus, you will remember,
allows the device to be addressed, or contacted, by the CPU, using a set of Input/
Output (I/O) addresses. There are also pathways for things like interrupts, direct
memory access (DMA) channels, and clock signals.
It is really pretty easy to tell what type of expansion bus you are using, just
by looking at the motherboard. As you have probably figured out by now,
I am big time into the history of computing, and it is never a bad thing to
know where the industry has come from. Some of these you may never see,
unless you go through a hardware museum, but it never hurts.
ISA 8-Bit Bus
Back in the early days, the expansion bus was only 8 bits wide and had a
blazing speed of 4.77 MHz. There were just eight interrupts (we will talk
about those in just a few pages) and just four DMA channels. By todays

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

System Bus Architecture

173

standards, this is slower than horse and buggy, but for the day it was blazing
fast. Since this was the very first PC Bus, and since it was designed by IBM,
the original makers of the IBM PC, they referred to this architecture as
Industry Standard Architecture (ISA). The 8-bit bus connectors are shown
in Figure 5.1, and an 8-bit bus expansion card is shown in Figure 5.2.
FIGURE 5.1

Eight-Bit bus connector

If you look carefully at the picture above, notice how wide those finger slots
were. We will be able to compare those with newer technology in just a second. Here is the type of card that took advantage of that slot (see Figure 5.2).
FIGURE 5.2

Eight-Bit bus expansion card

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

174

Chapter 5

System Bus Architecture

ISA 16-Bit Bus

As technology advanced, we went from an 8-bit to a 16-bit bus. This went
a long way to solving the biggest problem with the 8-bit busthe fact that
it was only 8 bits wide. The new ISA Bus had a 16-bit data path, and you also
may have heard it referred to as the AT Bus.
Look at Figure 5.3. Notice that the ISA Bus slots have a small bus connector behind the standard 8-bit bus. This connector provides several more
fingers, which are signal lines. The more signal lines available, the faster the
card can gather in information.
FIGURE 5.3

ISA Bus connectors

Motherboard

8-bit bus slots

So, what you have here is the old 8-bit slot with an add-on. The ISA Bus
also helped expansion by adding eight more interrupts and four more DMA
channels. It was quite easy to spot the kind of board that fit these new ISA
slotsthey looked like Figure 5.4.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

System Bus Architecture

FIGURE 5.4

175

ISA Bus expansion card

16-bit
connector

8-bit connector

If you look closely at the card, you will see that toward the front of the card
there is an 8-bit connector, separated from the read connector by a slot. This
architecture was really interesting because of compatibility. For example, if the
expansion card was an 8-bit card, it would run in either an 8-bit or a 16-bit
slot. It you had a 16-bit card, it would naturally run in the 16-bit slot for which
it was designed, but it would also run (albeit a lot slower) in an 8-bit slot. So,
pretty much everything was compatible with everything else.
Micro Channel Architecture (MCA)
About this time in the history of computing, the company that invented the
PC, IBM, was beginning to think the world was passing them by. Their share
of the market was steadily declining and they figured that they had to do
something to get it back. That something was the Personal System/2 (PS/2).
Along with the PS/2, IBM was introducing a new type of data bus called
Micro Channel Architecture (MCA). This bus was supposed to put the ISA
Bus out of business by utilizing a smaller connector with thinner fingers.
MCA was revolutionary because it was available in either 16-bit or 32-bit
versions. Secondly, it could have several Bus Mastering devices installed, and
the bus clock speed was about 25% faster than the old 8 MHz systems,
screaming along at 10 MHz. The really revolutionary part of the puzzle was
the way that you configured the expansion cards. In all the other bus technologies, the cards were configured by jumper settings, or by DIP switches.
With MCA, device configuration was done with software. These were the
first software-configurable expansion cards.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

176

Chapter 5

System Bus Architecture

This was an interesting concept, but it had some problems. First of all, all
device configurations were done from a single boot diskette that contained all
the information files for all the devices. When you made a change, the change
was not only written to the card, it was also written to this diskette. That diskette was the only diskette that knew what was in the system and how each
device was configured. At the time, I was working doing onsite hardware support. Whenever I ran into a PS/2 device (and some of them were servers) I
knew there was going to be trouble. I would ask for the configuration diskette,
and usually receive a blank look from the customer. It was then up to me to
find a PS/2 diskette and configure the entire system from scratch. Great idea,
but they forgot to take it that one extra step of either saving the configuration
files to a disk or making the devices able to provide configuration information
if asked by a setup program.

You could always tell an MCA card: They dont call IBM Big Blue for
nothing! Look at Figure 5.5.
FIGURE 5.5

MCA expansion card

Blue handle
Blue handle

As with most things that came out of IBM at the time, the MCA architecture was very proprietary. At a time when the buzzword was compatible,
IBM wasnt. In addition, IBM charged vendors who developed their own
expansion cards 5% of their gross receipts. Even way back then, margins
were slim on computer hardware, and this put the cost of MCA peripherals
out of sight.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

System Bus Architecture

177

EISA
Back in the late 80s to early 90s it was still a computer war out there. IBM was
selling PCs because they were IBM. The catch phrase at the time was, You
can never get fired for buying IBM. But there was competition, led by the
Gang of Nine. The Gang of Nine was made up of nine computer manufacturers that thought there had to be a better way than MCA to get faster speeds.
The Gang of Nine consisted of some of the top names in the industry at the
time: AST, Compaq, Epson, Hewlett-Packard, NEC, Olivetti, Tandy, Wyse,
and Zenith. They began to offer an alternative to MCA called Extended Industry Standard Architecture (EISA). For a while, EISA was popular in both 386
and 486 computers until about 1993 when PCI came along.
EISA had many of the same things going for it that MCA had, but it also
had compatibility with the older ISA board. Take a look at Figure 5.6 and
Figure 5.7.
FIGURE 5.6

EISA Bus connector

Now, one of the things you do not see in that picture is how deep the connector slots really were. They were about twice as deep as the old ISA slots
and 8-bits slots. Compatibility was done by staggering the finger slots. Look
closely at Figure 5.7 and you will see that some of the grooves are longer than
others.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

178

Chapter 5

System Bus Architecture

FIGURE 5.7

EISA expansion card

With this type of setup, if you were installing an 8-bit card, it would only
go so deep into the expansion slot. A 16-bit card would go as deep, but use
the back connector. An EISA card on the other hand, would slip all the way
to the bottom of the connector, making a 32-bit data path.
EISA Configuration

Are you familiar with Plug-and-Play hardware? Well, EISA was a precursor
to Plug and Play, and at the time, it was certainly a lot easier than other forms
of hardware installation.
Lets say that you were installing a new network card in an ISA-based
machine. Before you installed the network card, you had to check the computer to find out (at the very least) what interrupts were being used by other
devices. Then you configured the network card to use an interrupt that was
not being used by any other device, installed it, turned the computer on, and
ran the appropriate driver for the card. If you did your job right, the driver
would load and you had network connectivity. What usually happened, on
the other hand, was that you (okay, read that I) had guessed wrong and the
IRQ was already in use. This necessitated starting all over. Things changed
with EISA.
With an EISA Bus, on the other hand, you would take the top off the computer and install the EISA card in an EISA slot. The toughest part of the process was remembering what slot you installed it into. Anyway, after the card

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

System Bus Architecture

179

was seated, you turned the computer on, and as part of the Power On Self
Test (POST) the computer would figure out that there was something new,
different, and interesting going on inside. The computer would ask you to
configure the device, and you would use a program called EISA Configuration (EISA Config for short) to set the IRQ, DMA, and anything else you
needed to set. All this was done via the slot number, and the information was
then saved on the card. This made a technicians life remarkably easy,
because the EISA Config utility would even go out and check to find out
what settings were already being used. That way, you almost couldnt mess
it up.
The difference between configuring a machine with MCA and EISA was
that with MCA, you needed the diskette with the configuration utility for
that specific computer. Without the specific configuration disk, you reconfigured the whole machine. With EISA, you needed an EISA configuration
utility for that brand of computer. Sometimes, EISA configuration utilities
would even work across brands. So, if you carried around a diskette with the
Compaq EISA Config on it, you could configure all Compaq EISA machines.
There were other enhancements of EISA over ISA:


The CPU, DMA, and Bus Mastering devices could make use of a 32bit memory-addressing scheme.

The data transfer protocol that was used for high-speed burst transfers
was synchronous.

EISA supported intelligent Bus Master controllers for various peripherals.

EISA had better ways of handling DMA arbitration and transfer rates.

EISA made use of shareable interrupts.

EISA finally gave way to PCI. That is what the majority of systems on
the market are using today. Lets take a closer look at the new industry
standard bus.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

180

Chapter 5

System Bus Architecture

Real World Scenario

The preceding section contains a lot of material that you wont need to pass
the Server+ test, but it does give you a great background in the history of
hardware and how it has evolved. If there were certain parts that you should
pay close attention to, I would suggest looking over the material on the EISA
bus. EISA was a revolutionary architecture for its time, and there may even be
some EISA servers still trucking away in a production environment.
EISAs configuration utility was really a precursor to Plug and Play technology. The utility went out and found the devices that were installed into
the server and then helped you to configure those devices without the use
of jumpers or switches. All of that is taken for granted now, but for the
time, it was a radical concept.

Peripheral Component Interconnect (PCI)

ow that we have had a pleasant walk through memory lane, lets get
us closer to the present. When Intel released the Pentium processor, all of the
existing buses became instantly obsolete. Every bus up until this moment had
been of the 16-bit or 32-bit variety, and then along came the Pentium, which
was a 64-bit processor. Using a Pentium processor with a 16-bit or 32-bit
bus would be like pulling the engine out of a Ferrari and replacing it with
something from a Yugo. It just shouldnt be done and performance would
suffer greatly.
Peripheral Component Interconnect (PCI) works well with the current iteration of the processor. It can handle both a 64-bit and a 32-bit data path. It is
also processor independent, which means that it uses a form of Bus Mastering.

PCI Bus Mastering

Geek term alert! Bus Mastering is one of those terms that you tend to see
around every time you pick up a catalog or read a description of a new piece
of hardware. It is one of those terms that you may have heard and nodded
your head knowingly, all the while thinking to yourself, someday, I am going
to read a book where that term is really explained. That day is today.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Peripheral Component Interconnect (PCI)

181

Back in the early days of PCs it was up to the microprocessor in the computer to manage every byte that was moved along the data bus. It was up to
the microprocessor to read the byte from one device or from memory, decide
where that byte belonged, and then write the byte to the proper location. Soon
it became obvious that this was a whole lot of work that could be farmed out
to other devices. The microprocessor, for example, did not need to be handling
everything that went into and out of the expansion bus. After all, the microprocessor is supposed to be the manager of the operation, and all really good
managers know how to delegate responsibilities. Bus mastering is the result of
that delegation.
With Bus Mastering, the microprocessor does not have to be involved in
every transaction. It can delegate control to special circuits called bus controllers, and these bus controllers will direct traffic between different circuits. The
actual device that takes full control of the expansion bus is called a Bus Master.
The device that will end up receiving the data from the Bus Master is called the
bus slave. Some of the more politically correct systems may call the master and
the slave the initiator and the target.
So, the bus controller can manage multiple Bus Masters, and Bus Masters
take control of the actual expansion bus through a process called bus arbitration. Each type of bus has a protocol that is used for managing this arbitration
process. That protocol can be based in hardware or software, though it is usually
hardware-based.

PCI Bridges
Bus Mastering makes it sound like there is just one bus, and that is not even
close to the truth. The average PC has several buses and these are usually
operating at different widths and at different speeds. It is kind of a system
board designers hell. Somehow there has got to be a way to hook up all
those different types of buses together and get them to work in a cohesive
way. This really took the forefront when PCI was introduced, because
remember, PCI was designed to be processor independent.
The problem was solved with something called the PCI Bridge. Think
about what a bridge does in your world. It moves things from one location
to another over some kind of obstacle. That is just what a PCI Bridge does,
but it does it with data. The PCI Bridge moves that data from one system bus
to another system bus, and it is up to the bridge to handle all the gory details
of the transfer. This can include things like changing the data format and
protocols without making use of any outside hardware and software products. The bridge can be some form of standalone hardware, or it may just be
part of the chipset that makes up the PCs mainboard.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

182

Chapter 5

System Bus Architecture

PCI Bridges are really busy. In a typical system, for example, the bridge
can take moving information from the microprocessor bus to the high-speed
PC Bus and even to an old, outdated ISA compatibility Bus. PCI Bridges can
even link to other PCI Bridges to form a PCI-to-PCI Bridge, or a PPB.

How far can this go? PCI Bridges can be connected to other PCI Bridges up to
a maximum of 256 PCI Buses in a single PC. We will cover more on this when
we talk about Hierarchical Buses and Peer Buses.

Arbitration, 4-Bit Code, and Cache

Negotiation is the key! PCI uses a form of Bus Mastering that is arbitrated,
just like other types of expansion buses, but PCI has some advantages. The
first is that the PCI has its own bus command language that is a 4-bit code.
In addition, PCI Bus Mastering supports a secondary level of cache.
This is how it works: A PCI Bus Master board needs to take control of the
bus, so it sends the appropriate signal to the host. As soon as it receives confirmation to take control, the transfer is started. Every PCI board has its own
set of specific signals that are related to the slot the board is in. It can use
these slot-specific signals to ask for bus control and receive permission.
By using this system, there is a great deal of flexibility in the way priorities
are assigned to the computer. The system designer has the ability to change
the arbitration procedure to fit the needs of the expansion device rather than
the needs of the original, potentially outdated bus specification.
There are two special electrical signals that control Bus Mastering, the
Request (REQ#) and the Grant (GNT#). The process starts when the bus
master sends its Request signal when it wants to start controlling the bus. At
that point the circuitry called the central resource sends a Grant to the master
that gives permission to take control. Each PCI device gets its own dedicated
set of Grant and Request signals.

Hierarchical PCI Bus

Depending on the way the PCI Bridges are configured, you can have a server
that takes advantage of either a hierarchical PCI Bus or of a Peer PCI Bus.
With a hierarchical PCI Bus, the buses in the hierarchy operate concurrently.
That means that a PCI Master and a PCI target on the same PCI Bus can
communicate even if the other PCI Bus is busy. In this case, traffic between
devices on one bus is isolated from the traffic generated by all the devices on
the other bus, which should help to enhance system performance.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Peripheral Component Interconnect (PCI)

183

Other implementations of the Hierarchical Bus allowed devices on one

bus to communicate directly with devices on another. For example, National
Semiconductor released a PCI-to-PCI Bridge chip called the PicoPower Nile
II. This chip was designed to be used in high-end desktops, servers, and even
as part of multifunctional add-on cards. This chip provided additional PCI
Buses without system degradation.
The Nile II chip supported speeds of up to 120 MBytes/sec and allowed
for concurrent operation on both the primary and the secondary bus. This
bus hierarchy can support up to 256 PCI devices per bus, and there can be
a total of 256 buses. That totals over 65,500 devices.
In order to reduce the amount of time that the bus was waiting for other
devices, there were two special processes utilized. One was referred to as hidden arbitration and the other was bus parking. Hidden arbitration allowed
bus arbitration to go on while an initiator was performing a data transfer.
With bus parking, the PCI assigned the bus to a specific master, which was
usually the PCI Bridge. When the master had a request for the bus, the
request was granted immediately, since the bus was parked on it. The access
time to the bus in this case was as little as 60 nanoseconds (ns).
Take a look at Figure 5.8. This is a diagram of how a bridged PCI architecture works.
Bridged PCI architecture
Host Bus

Processors

Host-to-PCI
Bridge

133MB/sec

FIGURE 5.8

PCI Bus

Memory

Slots

PCI-to-PCI
Bridge

PCI Bus 133MB/sec

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Slots

184

Chapter 5

System Bus Architecture

You will notice that there is only one data path to get to the host bus.
Everything has to go through the PCI-to-PCI Bridge to reach the primary bus
and then through the Host-to-PCI Bridge. While this method does provide
for a great number of devices, there is no load balancing capability.
Lets see what it is like with the Peer PCI Bus.

Peer PCI Bus

The other way that a primary and secondary bus can connect to the host is
with a Dual Peer PCI Bus. Take a look at Figure 5.9 to see how the Dual Peer
is configured.
FIGURE 5.9

Dual Peer PCI Bus

Host Bus
540MB/sec

Processors

Slots

Bridge

Hostto-PCI

Bridge

133MB/sec

Memory

133MB/sec

Secondary PCI Bus

Slots

Primary PCI Bus

PCI-to-EISA

Bridge

33MB/sec

Slots

EISA Bus

You will notice that, in this case, the two PCI Buses are linked independently to the processor bus using two Host-to-PCI Bridges. Since there are
two independent buses, there can be two Bus Masters transferring data at the
same time, giving more overall throughput and a higher bandwidth. This is
especially useful if you have a server with two or more peripherals that are
bandwidth intensive. If you split the peripherals between the two buses, you
are in effect creating load balancing.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Peripheral Component Interconnect (PCI)

185

If you are using a server that makes use of the Peer PCI Bus, there has to
be configuration of the Input/Output (I/O) subsystems. The load balancing
configuration should be taken into account even before the initial system
setup and configuration takes place.
This bus balancing is accomplished by actually balancing the I/O bandwidth for each bus. This should produce the optimal performance on a system. This will work great with Peer PCI Buses, but it may not work as well
with a bridged PCI system. Here are some recommendations on when to do
load balancing.


If you are using a bridged architecture, load balancing is not recommended. With a bridged architecture make sure the primary bus is the
first one that is populated.

For Dual Peer architecture, load balancing is recommended. If the I/O

throughput is high, then the workload should be spilt between the
buses.

If your Dual Peer architecture also makes use of PCI Hot Plug slots,
there is going to be some tradeoff between high availability and high
throughput.

Bus Balancing
Here are some guidelines on how to balance a PCI load:


If you have several network or array controllers, make sure they are
split between the buses.

If you are installing an odd number of controllers, like two network

interface cards and a single drive array controller, split the two network cards between the buses. Network controllers use more bandwidth than drive controllers, so split the workload between the buses.

Avoid putting two network cards on the same bus, unless both buses
already have a network card installed. It is better to have a system that
has a dual-port network card on each bus, rather than to have two
individual network cards on each bus.

So, how does the processor bus know when the network cards need attention? PCI Buses do use interruptsthey just use them a little differently.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

186

Chapter 5

System Bus Architecture

PCI Interrupts
PCI is a self-contained expansion bus. The interrupts that would normally be
set at the card level are managed at the expansion-slot level by the software
that drives the peripheral devices. With PCI, there are four level-sensitive
interrupts that have interrupt sharing enabled, and these can amount to up
to 16 separate interrupts when examined as a binary value. The PCI specification does not define what the actual interrupts are for each slot or even
how they are to be shared. All of that design relationship is left up to the person who is designing the expansion device. That means that these details are
usually not handled at the hardware level, as was the case in the earlier architectures. With PCI devices, the software device driver for the board handles
the interrupt configuration. These interrupts are really independent in a way,
because they are not synchronized with any of the other bus signals and so
they can be activated at any time.

PCI Expansion Slots

Earlier in the chapter you saw examples of ISA, and EISA slots as well as an
MCA card. Take a look at Figure 5.10.
FIGURE 5.10

PCI Bus connector

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Peripheral Component Interconnect (PCI)

187

As you can see, the finger slots in the bus are very small, and packed very
closely together. These expansion slots are usually white, and they are
divided into two sections.
There are two different kinds of PCI expansion slots and the voltage the
slots use differentiates the versions. One of the types uses +5 volts DC to
power the expansion card, while the lower-voltage model uses 3.3 volts.
When you look at the connector for the buses, the only differences are the
positioning of the blocker in each connector. This blocker, or key, keeps the
3.3-volt card from being plugged into a 5-volt slot.
Now, you have been wondering why I spent all that time covering some
of that other stuff on PCI Bridges and what have you. Well, when we talk
about expansion slots, one of the first questions that comes to mind is how
many can you have. The answer to that is, It depends. As you start stuffing
more and more stuff in a smaller and smaller space, something has to give,
and it is usually the electrical effects inside any given system. Because PCI
operates at a high bus speed, it is especially susceptible to high frequencies,
radiation, and other forms of electrical interference. The current standards
for a local bus limits to three the number of high-speed devices that can be
connected to a single bus.
If you paid close attention there, you notice that the standard calls for just
three devices, not three slots. Most local bus systems now have their video
display built into the motherboard. That circuit counts as a local bus device,
so, if your PC has video on the motherboard, you can use two local bus
expansion slots.
The limit of three devices comes from speed considerations. The bigger or
larger the bus, the more connectors there are. More connectors means that
any signal placed on a circuit will degrade more quickly, and the only way
to beat the degradation is to start with more signals. Somewhere, someone
had to draw the line, and the line was drawn at three devices.
While it seems that three devices may be limiting, it is not. Remember our
discussion of PCI Bridges? Well, since the three-device limit is per expansion
bus, the PCI Bridge allows multiple expansion buses to be linked together.
Each of these will use its own bus control circuitry. While this may sound
complicated, it is one of those things that doesnt really make any difference.
After all, as long as it works, that is all that counts, and the design is all in
the chipset.
Is there a way you can use this technology not only to increase the performance, but also to increase the availability?

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

188

Chapter 5

System Bus Architecture

PCI Hot Swap

If you remember way back in the book to Chapter 1 when we discussed
RAID technology, I also touched on things like a hot-swappable drive or a
hot-pluggable drive. In case you missed it, here is the premise. It used to be
that servers and networks were fragile and everyone took it for granted that
they went down occasionally and that was a fact of life. That was then, this
is now. Now, the server is an integral part of the operation of any company.
Companies can lose hundreds of thousands of dollars an hour in productivity if their servers are down. The goal, as far as management is concerned, is
to keep the servers up.
Now, you and I are networking professionals. We both know that no matter
how carefully you plan, no matter how carefully you design, no matter how
many precautions you take, life happens and things like components fail. This
used to be expected and it was the norm to schedule an outage, take a server
down for a couple of hours during the slow periods, and then have it back up
and working for Monday morning. More and more shops are working 24 hours
a day, seven days a week, so any downtime needs to be severely limited.
Hot swap, in theory, is simple. Why take the entire server down when all
you have to replace is one device or one card? Why not just replace the card
and be done with it? That is the basis for hot swapping. Basically, there must
be the ability to replace a plug-in board in the system while the power is on
and active. The goal is to have the maintenance people just remove the failed
component and insert a replacement. If any other procedure is completed, it
can create an operator error.
Now, if you are talking about hot swapping a hard drive immersed in a
RAID array, that is one set of challenges. Electronically, though, it is uncomplicated. You unplug the drive from the electronics, unplug the connection
from the drive controller, and remove the drive. Then, you insert the new
drive, hook up the controller cable, and apply power. The power surge
would be similar to turning a computer on and the controller would be notified the new drive is online. With PCI, however, you are removing a card
from a live slot, and replacing a card in a live slot. I can see there would be
some serious challenges.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Peripheral Component Interconnect (PCI)

189

When you are talking about hot swapping PCI devices, there are two sets of
specifications, the Hot Swap PCI Specifications and the CompactPCI Hot Swap
Specifications that are managed by the PCI Industrial Computer Manufacturers Group (PICMG). The two standards are very similar and they differ in only
a couple of areas. For example, with the PICMG Specifications, the backplane
that the device plugs into is passive, and all the logic is contained on the
adapter card. This same logic is used to power up the adapter card.

Making it easier for the developers is the fact that the devices are controlled
by software. It is up to the system software to provide the smarts for this whole
process to work.

System Software and Hot Swapping

When you start looking at this kind of awareness, it must be instituted at the
operating system or at an application. Either way, that piece must communicate with the software components running on a card. The card and the application can work together to shut things off and make it possible to replace the
PCI card. The PCI Hot Swap Specifications lays out six different classes of
software that are Hot Swap compliant. The minimum requirements allow for
just live insertion and live extraction. This is done with two levels of software.
The bottom layer of the software has three hot swap performance grades that
are applied in either a specific use (proprietary) or a general use category.
Specific use software is implemented when the operating system does not
specifically support PCI Hot Swapping. The application software that is
developed in this category cannot necessarily be moved to other platforms.
General use software, on the other hand, is fully compliant with, and usually integrated into, the operating system. This provides for the widest support of platforms and also gives the programmers of board drivers a specific
set of application programming interfaces (API) to use. When it is put
together it looks like Figure 5.11.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

190

Chapter 5

System Bus Architecture

FIGURE 5.11

Hot swap software implementations

Application

Device Services Requests/Grants

Hot plug
service

OS Calls Service

OS

Primary Command Interplay

Hot plug system driver(s)

Hot plug system driver(s)

PCI hardware

Within the general use and the specific use categories are three levels that
define how the live-insertion capability is carried out. These levels are Basic
Hot Swap, Full Hot Swap, and high availability.
Basic Hot Swap The end user must tell the operating system that a card
is going to be inserted or removed. This is usually done from the system
console.
Full Hot Swap This category adds to the functionality of Basic Hot
Swap. In this case, there is a microswitch added to the cards injector/ejector mechanism. This way, the technician does not have to tell the operating system that the change is about to occur. When the card is installed or
removed, the switch changes the electrical configuration and gives the OS
a warning that the process is about to occur.
High availability This level provides the greatest functionality for reconfiguring software while the system is running. This allows for on-the-fly
reconfiguration of both the hardware device and the software components.
In this case, the operating system itself can sense when a card has failed, and
the OS will bring a previously installed replacement card online to assume
the duties of the failed device.
Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Peripheral Component Interconnect (PCI)

191

Hot Plug PCI

The goal of Hot Plug PCI is the same as the goal for hot swap basically,
keeping a system that needs to have high availability available. In the case of
the revision 1 of the PCI Hot Plug Specifications, giving the chipset on the
device a way to deal with various software requests does this. These requests
include things like these:


The ability to choose to isolate a card from the logic on the system
board.

The ability to choose to remove or apply power to a specific card

connector.

The ability to choose to turn on or turn off an Attention Indicator that

is associated with a specific card. This attention indicator can then be
used to draw the users attention to the connector.

With Hot Plug PCI, the user cannot remove or install a PCI card without
first telling the software. Once the software has been notified, it performs the
steps necessary to shut down the card connector so the card can be removed
or installed. It is up to the operating system to visually let the end user know
when it is all right to install or remove the card.
The advantage of Hot Plug is that you can use any PCI card in the system.
Changes are needed to the chipset, the system board, the operating system,
and the drivers.

I2O: Intelligent Input/Output

Intelligent Input/Output (I2O) is designed to provide a dedicated I/O processor on certain devices, like a server motherboard or even on network
cards or disk controllers. The standard is going to provide an I/O architecture that is independent of both the hardware device that is being controlled
and the host operating system.
I2O is made up of three software layers: the OS Services Module (OSM), the
I2O Messaging Layer, and the Hardware Device Module (HDM). The OSM is
designed to manage the communication between the host CPU operating system
and I2O Messaging Layer. This layer is unique for each operating system and
also for each device class. The I2O Messaging Layer is in the middle, handling
communications between the OSM and the HDM. The Messaging Layer frees
the device manufacturer from having to develop different drivers for the same
device. Finally, the HDM passes messages between the peripheral device and the
I2O Messaging Layer. The HDM has to be unique for each device but, if you
have several of the same kinds of devices, only one HDM is required. The HDM
is independent of the host CPU operating system.
Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

192

Chapter 5

System Bus Architecture

I2O can provide the server with the following benefits:



Provides an I/O architecture that is both extensible and independent of

specific hardware or operating systems.

It improves the throughput of the server, because the I/O of the peripheral
is removed from the CPU.

Provides an industry standard architecture.

Can increase fault isolation and recovery, which works to provide for
higher availability.

Provides for direct management of the I/O.

AGP
In each of the previous sections we have been talking about throughput,
especially when it comes to network interface cards and disk controllers. In
each of these sections we stressed how important it was to off-load the mundane tasks from the processor, in effect, giving it more time for the serious
processor tasks.
This section is going to take a somewhat different tack, concentrating on
video. Now, this is not necessarily a topic I normally associate with servers,
because usually servers are servers and high-definition graphics is not all that
important.
AGP is short for Accelerated Graphics Port. It is an interface based on
PCI and designed for the throughput demands of high demand video like
3-D graphics. Rather than using PCI for graphics data, AGP has a dedicated point-to-point channel to directly access main memory. AGP runs at
66 MHz over a data channel that is 32 bits wide, providing bandwidth of
266 MBytes/second. This compares to the PCI bandwidth of 133 MBytes/
second. In addition, there are two optional, faster video modes, providing
throughputs of 533 MBytes/second and 1.07 GBytes/second. AGP can support this kind of throughput by storing some of the 3-D textures in main
memory rather than in video memory.
Why would you install AGP on a server? Well, if there are 3-D applications that you have to run on the server, AGP will help off-load some of the
work that is placed on the CPU. If there is a 3-D application running, the
CPU (without an AGP graphics controller) is responsible for performing all

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Peripheral Component Interconnect (PCI)

193

those intensive 3-D calculations. The graphics controller can process the texture data and the bitmaps. At this point, the controller has to read information from several different textures and then average the information into a
single pixel on the screen. While this calculation is being performed, the pixel
is stored in the memory buffer. Since the textures are very large, there isnt
room in the video cards buffer. AGP overcomes this shortcoming by storing
the image in main system memory.
When AGP wants to access the texture data, it uses a process called Direct
Memory Execute (DIME). DIME connects the systems main memory to the
AGP/PCI chipset.
So, should you be looking for an AGP controller in your server? If your
server is going to be physically running some 3-D applications, it may be
something you want to look at. However, you should know that several published studies question whether there really is a performance increase over
using just a PCI video card. If your system is going to be making use of AGP,
you should definitely add more memory to the server to provide the extra
memory that the video subsystem needs.

Real World Scenario

This is another of those sections where the information provided will help you
understand the makeup of your current server. When you start talking about the
different methods of PCI Bridging, chances are this is not something you are
going to go out and add to a current server. Your high availability, high utilization
servers will have these things configured at the factory.
PCI really went a long way toward removing the mystery of component
installation. Before PCI came along, installing a new component or reconfiguring the system bus was not for the faint-hearted. You needed to have
lots of up-to-date information, and even then, things could get somewhat
confusing if the information was not exactly right. At that point, you had
to do some troubleshooting by trial and error.
Early bus architecture also limited the expandability of systems. Since IRQs
could not be shared and the number of usable IRQs was limited, you were
extremely limited in the number of additional components you could add to a
server. In some cases, you ran out of IRQs before you ran out of slots!

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

194

Chapter 5

System Bus Architecture

The System Bus and System Performance

All the network administrators that I know really hate it when they
hear the complaint, Gosh, the network is slow today!
In this chapter, I have laid out several different technologies that can help
you provide both performance and high-availability solutions. As you saw at
the beginning, the early motherboards bus technology was speed limiting,
not only in processing power but also in moving the information from the
processor back to the user who requested it. PCI helped to change that.
Each of the technologies that we have talked about has stressed the same
philosophy: Take the mundane calculations away from the processor and let
something else handle it. That way, the processor is freed up to do other
things. This, in turn, speeds up performance.
When you design your server, pay close attention to the types of subsystems
that are present, and be sure to take full advantage of them. Also understand
that for each of the performance enhancing technologies that you opt to have
in your server, there are going to be trade-offs. Usually that trade-off will come
in the form of how large a check you will have to write to pay for the server.

Upgrading Peripheral Devices

system peripheral is an add-on component that performs a service not

typically associated with the main server computer. For example, a tape
backup device is considered a peripheral, as is a CD player, a CD writer, a
scanner, a multi-port serial device, and so on. Managing peripherals becomes
a big part of a server admins job when its time to upgrade but after that, provided the peripheral is behaving OK, the device is basically a set-it-and-forgetit thing.
Youll be surprised how often you entertain the notion of upgrading
peripherals. For example, in the Digital Linear Tape (DLT) backup device
category, you mightve inherited a server that was able to put 40 MBytes/
minute of compressed data onto a DLT tape but today you want to replace
it with a tape drive thats capable of 80 MBytes/minute. So a peripheral
upgrade is in order. Or maybe you want to upgrade that old 4x CD-ROM
with a nice new 48x device.
There are some very rare occurrences in which you can upgrade a card in
a peripheral rather than having to upgrade the entire peripheral, but those
times are few and far between.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Upgrading Peripheral Devices

195

Realize too that you can often get a few extra miles out of an older peripheral by simply upgrading its system BIOS. This may not be possible with all
devices, but many of them have the ability to have their firmware updated to
make them compatible with newer operating systems.

Verifying Appropriate System Resources

When considering upgrading a peripheral, you must take certain key physical
elements into consideration. Well use this section to deal with some items that
you must think about (and document on a checklist) before purchasing and
installing a peripheral upgrade.
Expansion slots The first thing to verify is what type of slot the new
device will occupy (ISA, EISA, PCI, etc.) and that you have an open slot
of the appropriate type to receive the new device. This determination will
be meaningful to you when youre considering upgrade slot-based devices
such as RAID array controller cards, SCSI adapters. and things like that.
Some devices require a slot in the computers frame as well as a cable to
a card mounted in the motherboards chassis. A CD-ROM has this kind
of configuration requirement. If you dont already have a card in the computer that can accept the CD-ROMs cabling, then youll have to plan
accordingly.
Having enough slots in the computers frame is also an issue when considering peripheral upgrades. Some computers are filled to the brim with
peripheral devices and can accept no more unless, of course, you take out
an old peripheral and replace it with the new, which is typically what
youll do.
Some servers have quick disconnecting slots that allow you to easily Plug
and Play peripherals in and out of the system. The server chassis designers
are beginning to create very smart admin-oriented designs that allow you
to get a peripheral upgraded in a hurry.
Interrupts One of the bigger problems youll run into is having enough
interrupts (IRQs) to handle the replacement. Again, if youre simply swapping out, you may not have an issue. But what if youre swapping out an
ISA-based backup device with a SCSI-based unit but you find that you dont
have any IRQs available for the new device? Well, then youve got to look
at getting rid of things you dont need so that you can free up an IRQ. For
example, COM2 utilizes an IRQ that you may not need. Go into the
servers BIOS, disable the COM2 port, and youve freed up an IRQ that you

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

196

Chapter 5

System Bus Architecture

can use for your device. In situations where youre not sure about the IRQs,
figure out what IRQs are in use first, then youll know whats available for
the new device.
Direct Memory Access Direct Memory Access (DMA) provides a way
that data can be transferred from a device to system memory or vice-versa
without having to go through the CPU, thus freeing up CPU cycles. You
set up a DMA channel for the data to go through. DMA isnt heavily used,
but it should be used more than it is. When purchasing new peripheral
gear, check the products documentation to see if it can use DMA, then
decide which DMA channel youd like to set up for the device.
Cabling Cabling is a huge issue for external SCSI devices. Youll have to
look at the back of the computer to determine what type of connection the
internal SCSI adapters external port has. Next you determine what kind
of SCSI connection the new device is expecting. Finally you purchase a
cable that matches the configuration. For example, suppose that youre
going to purchase an Ultra-SCSI device but plug it into a SCSI II external
port on the computers SCSI adapter. Youll need a SCSI II-to-Ultra-SCSI
cable. Youll want to make sure which side is male and which is female as
well, before you go looking for the cable. You can buy adapters that fit
onto a SCSI cable to make the cable work with different SCSI versions. I
think youre setting yourself up for data transfer problems if you purchase
an adapter because it could work loose and cause you some problems that
may be hard to diagnose.
Note that you might want to go into the SCSI adapters BIOS to tweak it so
it works with the new device. Check your SCSI adapters documentation
for more information on adjusting BIOS settings.
Power Some peripherals require a power socket and are separately powered from the computer. Be aware of this before you buy so that if youre
lacking enough power sockets where you want to place this peripheral, you
can get the electrical work done before the peripheral comes in. For example, some backup tape devices require a substantial power supply and youll
have to address the power needs before the gear can be put into production.
When your new gear comes in, read its documentation thoroughly and be
sure you understand how to install and configure the device. Lots of times
its easy to get in a hurry and think that you dont need to bother with reading the documentationbut its always worth your while to be sure you read
and understand how the device is supposed to interplay with your system.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Summary

197

Performing an Upgrade Checklist

Be sure to establish some sort of upgrade checklist that you can use to validate
that youve gone through all of the procedures needed to verify that youll have
a successful implementation of the new device. Your checklist should consider
the resource characteristics discussed above as well as updated driver downloading, checking the FAQs and readmes for current information, and other
pertinent items.

Summary

In this chapter we completed the hardware portion of the book. We

have looked at all the various hardware subsystems, from the disk drives,
to the processor, down to the bus. Here we looked at ways to increase the
performance and availability of the PCI subsystem, which is going to be
found in all your servers today.
In the next few chapters we are going to look at how the server can be
actually used. We will begin by looking at some management protocols and
then move into various server roles. So, now that you know how to build
your server, it is time to figure out how to use it!
We talked about the upgrading of peripheral devices and some of the
things that you have to be cautious about relative to an upgrade. You must
make sure you have ample chassis and motherboard slots to house the new
gear. Youll also want to be cognizant of IRQ and DMA settings and to make
sure you have enough power and the right kind of cabling for the new device.

Exam Essentials
Know the basics of PCI bus mastering PCI Bus Mastering is a way for
the motherboard bus to improve performance by directing signals directly
to the components. This is one way of making sure the CPU is involved in
only those transactions that it really has to act on. If the workload of the
CPU is eased, your server should experience better performance.
Know the basics of PCI hot swap or PCI hot plug PCI Hot Swap
means that you can remove a bad component and replace it without shutting off the server. PCI Hot Plug means that you can add a component
without taking the server out of service.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

198

Chapter 5

System Bus Architecture

Know the basics of a hierarchical and peer PCI bus With a hierarchical
PCI Bus, the buses in the hierarchy operate concurrently. That means that
a PCI Master and a PCI target on the same PCI Bus can communicate even
if the other PCI is busy. With a peer to peer PCI Bus there are two independent buses. This means that there can be two Bus Masters transferring
data at the same time, giving more overall throughput and a higher bandwidth. This is especially useful if you have a server with two or more
peripherals that are bandwidth intensive. If you split the peripherals
between the two buses, you are in effect creating Load Balancing.
Know what interrupts are and how the system uses them Interrupts
(IRQs) are the way components get the attention of the CPU.
Know that EISA is a form of system bus; know how the architecture of the
system bus can affect server performance The architecture of the system
bus will determine how much information can flow to various components
at any given time. The faster the bus, with the appropriate components, the
better the performance should be.
Be able to upgrade a variety of devices Know and understand the
complexities and nuances of installing upgraded peripheral devices.

Key Terms
Before you take the exam, be certain you are familiar with the following terms:
Accelerated Graphics Port
address bus
Basic Hot Swap
blocker
bus
bus arbitration
bus controllers
Bus Master
Bus Mastering
bus slave
clock signals

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Summary

data bus
direct memory access (DMA
Direct Memory Execute (DIME)
EISA Configuration
expansion slot
Extended Industry Standard Architecture (EISA)
external bus
Full Hot Swap
Grant (GNT#)
Hardware Device Module (HDM)
Hierarchical PCI Bus
Host-to-PCI Bridge
Hot Plug PCI
Hot Swap
I2O Messaging Layer
Industry Standard Architecture (ISA)
Input/Output (I/O)
Intelligent Input/Output (I2O)
interrupt requestor (IRQ)
interrupts
load balancing
Micro Channel Architecture (MCA)
OS Services Module (OSM)
PCI Bridge
PCI Hot Plug
PCI-to-PCI Bridge
Peer PCI Bus

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

199

200

Chapter 5

System Bus Architecture

Peripheral Component Interconnect (PCI)

Personal System/2 (PS/2)
Request (REQ#)

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

201

Review Questions
1. What was the first computer bus referred to as?
A. EISA
B. MCA
C. ISA
D. I2O
2. EISA was referred to as which of the following?
A. An 8-bit bus
B. A 12-bit bus
C. A 16-bit bus
D. A 24-bit bus
E. A 32-bit bus
3. If you want to do PCI load balancing, what will you need to have?
A. A Peer Bus
B. A Hierarchical Bus
C. Hot Swap devices
D. Hot Plug devices
4. Which version of Hot Swap PCI requires users to notify the operating

system that they are about to take a device out of the system?
A. Basic Hot Swap
B. Full Hot Swap
C. High availability
D. All of the above

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

202

Chapter 5

System Bus Architecture

5. How many devices can be part of the original PCI Bus?

A. Five
B. Four
C. Three
D. Two
E. One
6. If you are using PCI load balancing and you want to install two network

interface cards and a single drive array controller, how would you plan
to install them in a Dual Peer Bus configuration?
A. Put both of the NICs on one bus and put the drive array controller

on the other.
B. Put both NICs on the master PCI Bus and the drive array controller

on the slave PCI Bus.

C. Put a NIC on each of the peers, and add the drive array controller

to either one.
D. There cannot be more than one NIC in any server.
7. What is Bus Mastering?
A. All transactions are sent directly to the processor.
B. All transactions are sent directly to memory.
C. All transactions directed to the disk array controller are directed to

the PCI controller.

D. The microprocessor does not have to be involved in every transaction.
8. AGP is a type of which of the following technologies?
A. Disk Array
B. Video
C. PCI Hot Swap
D. PCI Hot Plug

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

203

9. AGP makes use of what type of memory?

A. L1 cache
B. L2 cache
C. Main memory
D. High memory
10. With a PCI Bus, how are interrupts configured?
A. Software
B. Jumper
C. DIP switch
D. Interrupts are hard coded to the expansion slot.
11. What is I2O designed to do?
A. Make use of main memory
B. Make use of L2 cache
C. Provide a dedicated I/O processor
D. Provide a way of accessing load balanced PCI devices
12. Johann is replacing his old 8mm tape drive with a new five-cartridge

DLT tape changer. Both devices are SCSI. Now the computer wont
boot to the NOS and Johann is getting a SCSI IRQ conflict error even
though he verified that hes using the same IRQ as the old backup
device. What could be the problem?
A. New device isnt terminated.
B. Device is trying to use six IRQs.
C. PCI bus is autodetecting the wrong IRQ.
D. New devices BIOS hasnt been enabled.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

204

Chapter 5

System Bus Architecture

13. What is an advantage of Hot Plug over Hot Swap?

A. Hot Plug can use regular PCI cards.
B. Hot Plug PCI cards are cheaper than Hot Swap cards.
C. There are no differences.
D. A, B
14. Ahmed has a server to which hes adding a second PCI NIC card in

order to dual-home it. The card went in OK and didnt seem to

complain but now the network connectivity is acting erratic. What
could be the problem?
A. I/O address conflict
B. IRQ conflict
C. DMA address conflict
D. PCI bus disallowing NICs IRQ
15. What are the two types of PCI Bridges?
A. PCI-to-CPU
B. PCI-to-PCI Bridge
C. Host-to-PCI Bridge
D. PCI-to-Host Bridge
16. If PCI Bridges are connected to other PCI Bridges, what is the maxi-

mum number of PCI Buses?

A. 512
B. 256
C. 128
D. 64
E. 32

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

205

17. What are the three special circuits that make use of Bus Mastering?
A. Bus driver
B. Bus controller
C. Bus tool kit
D. Bus master
E. Bus slave
18. MCA architecture is designed for what types of computers?
A. ISA
B. EISA
C. 386 and 486
D. PS/2
E. AT
19. What made EISA obsolete?
A. PCI
B. MCA
C. ISA
D. AT
20. If AGP wants to access texture data from memory, what is the name

of the process it uses?

A. NICKEL
B. DIME
C. PENNY
D. QUARTER

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

206

Chapter 5

System Bus Architecture

Answers to Review Questions

1. C. The first computer bus was introduced by IBM and was called the

Industry Standard Architecture (ISA).

2. E. EISA was a 32-bit bus.
3. A. You can do PCI load balancing with a Peer Bus.
4. A. The only version that requires user intervention is Basic Hot Swap.
5. C. The original PCI Bus was limited to three devices.
6. C. In this case, you would place one NIC on each of the peers and then

add the controller to either of the peers.

7. D. With Bus Mastering, the microprocessor does not have to be

involved in every transaction.

8. B. AGP is a way of working with 3-D video.
9. C. AGP uses main memory.
10. A. Interrupts are configured by software.
11. C. I2O provides a dedicated I/O processor on a variety of types of

devices.
12. B. While answer A is somewhat likely, its probably not going to be an

issue with todays SCSI devices. Most of the current batch of devices
autoterminate so you dont run into the old SCSI termination issues of
yore. On the other hand, if youre trying to plug in a tape changer that
thinks each tape device needs its own IRQ, then youre subbing out a
tape backup device that used only one IRQ with a device that needs six
(one for itself and five for the other drives).
13. A. The advantage of Hot Plug is that you can use any PCI card in the

system.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Answers to Review Questions

207

14. A, B. Most likely the card has defaulted to IRQ 5 and the other NIC

in the box is also set for IRQ 5. It could be too that both NICs are
using the same I/O address. You could have a server boot up and seem
to work OK with a scenario such as this and find it very difficult to
troubleshoot. Typically the server NOS errors logs will pinpoint the
problem.
15. B, C. The two types of PCI Bridges are PCI-to-PCI Bridge and Host-

to-PCI Bridge.
16. B. PCI Bridges can be connected to other PCI Bridges up to a maxi-

mum of 256 PCI Buses.

17. B, D, E. The processor can delegate control to special circuits called

bus controllers, and these bus controllers will direct traffic between
different circuits. The actual device that takes full control of the
expansion bus is called a Bus Master. The device that will end up
receiving the data from the Bus Master is called the bus slave.
18. D. MCA was designed for PS/2 computers.
19. A. The PCI Bus made the EISA Bus obsolete.
20. B. If AGP wants to access the texture data, it uses a process called

Direct Memory Execute (DIME).

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Chapter

Server Functions
SERVER+ EXAM OBJECTIVES COVERED IN
THIS CHAPTER:

1.1 Conduct pre-installation planning activities.


Plan the installation.

Verify the installation plan.

Verify hardware compatibility with operating system.

Verify power sources, space, UPS, and network availability.

Verify that all correct components and cables have been

delivered.

1.2 Install hardware using ESD best practices (boards, drives,

processors, memory, internal cable, etc.).


Mount the rack installation.

Cut and crimp network cabling.

Install UPS.

Verify SCSI ID configuration and termination.

Install external devices (e.g., keyboards, monitors,

subsystems, modem rack, etc.).

Verify power-on via power-on sequence.

2.1 Check/upgrade BIOS/firmware levels (system board, RAID,

controller, hard drive, etc.).

3.2 Add Processors.


On single processor upgrade, verify compatibility.

Verify N 1 stepping.

Verify speed and cache matching.

Perform BIOS upgrade.

Perform OS upgrade to support multiprocessors.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Perform upgrade checklist, including: locate/obtain latest

test drivers, OS updates, software, etc.; review FAQs,
instructions, facts, and issues; test and pilot; schedule
downtime; implement ESD best practices; confirm that the
upgrade has been recognized; review and baseline;
document upgrade.

3.5 Upgrade BIOS/firmware.



Perform upgrade checklist, including: locate/obtain latest

test drivers, OS updates, software, etc.; review FAQs,
instructions, facts, and issues; test and pilot; schedule
downtime; implement ESD best practices; confirm that the
upgrade has been recognized; review and baseline;
document upgrade.

4.3 Set SNMP thresholds.

4.4 Perform physical housekeeping.

4.5 Perform hardware verification.

4.6 Establish remote notification.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

ow that we are all done with the hardware aspects of a

server, we can start talking about the business case that the server was purchased to perform. Business case? What the heck is this, a business management test or a networking test? Actually, it is a little of both. You see, it used
to be that local area networks (LANs) were installed and configured and
then the IT department told people how they could use the network. That
was the way it used to be. Now, however, servers (and networks for that
matter) are brought online to serve the needs of the customer, and that usually means answering some specific business need.
First we will look at the grouping of servers that serve the business needs
of the IT department. These servers can help the IT department to troubleshoot problems, as well as proactively respond to problem areas before they
happen. That is one of the appropriate ways to use management software.
Next we will look at all the different roles that a server can fill in the network environment. There are a bunch of roles and, depending on the size of
the server, some boxes can, and will, serve multiple functions. We have a lot
to do in this chapter, so we might as well get started!

For complete coverage of objective 1.1, please also see Chapter 9. For complete coverage of objective 1.2, please also see Chapters 1, 7, and 8. For
complete coverage of objectives 2.1, 3.5, and 4.4, please also see Chapter 8.
For complete coverage of objective 3.2, please also see Chapters 3, 8, and 9.

Management Protocols

Now, if you are managing a real live computer network, I dont have
to tell you that there are two things you want. You want more information

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

212

Chapter 6

Server Functions

about what is happening on the network infrastructure and you also want to
be able to figure out what the heck is going on at the desktop level. That is
why it is important to understand both SNMP and DMI.
Lets get the big picture first and then drill down to the individual desktop.

Simple Network Management Protocol (SNMP)

The concept behind Simple Network Management Protocol (SNMP) is
really quite simple. You need more information on your network, and it
would be wonderful if all of the devices and all of the software that make up
your network infrastructure could tell you what was going on. That is
exactly what SNMP is designed to do.
SNMP is one of the protocols that make up the TCP/IP protocol suite, but
it also works with IPX networks. We will be talking about some of the other
TCP/IP protocols in this chapter when we talk about things like DHCP,
DNS, and FTP. SNMP can be thought of as the micromanaging middle manager or the pointy-haired boss of the network management. This is the person that requires you to keep track of everything, even though he or she will
never ask for it.

For the official version of SNMP, you can see Request For Comments (RFC) 1157.

SNMP in the wild is made up of an SNMP Agent and an SNMP Manager.

For security and management purposes, all of the SNMP Agents can be part
of a community. To see how it works, lets start with an SNMP Agent.

SNMP Agent
Think of a secret agent, gathering in all the information that can be gathered,
never knowing what will be useful and what wont. This is an agent who
never makes decisions on the quality of the data; the agent only gathers the
information in and gets it ready to pass on to the person who is the next in
line. That is what an SNMP Agent can do.
An SNMP Agent can be a hardware device, like a router or a switch or it
can be a software process, like a DNS Server process or an Internet e-mail
connector. No matter what it is, it shares certain characteristics with other
SNMP Agents. First of all, it is providing some kind of service. Secondly, it
is a service that is important to the health of the networkotherwise you
wouldnt want to know about what was going on.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Management Protocols

213

An SNMP Agent keeps track of events. Depending on the type of Agent,

an event could be something like a logon or logoff, it could be the number of
packets processed, or it could be the number of pages printed by a certain
printer. In any case, the event that is being kept track of has been defined by
something called a Management Information Base (MIB).
Now, when a device or process comes online, it checks this MIB to see
what it is that the device is supposed to track. There are various levels of
MIBs, some tracked by standards agencies, and some defined by software or
hardware. Depending on the device, a standard MIB may be enhanced to
track a particular process carried out by a particular device.

SNMP Managers
Like all good managers everywhere, an SNMP Manager is interested in
information. The information that it is interested in can be user configurable
on hardware devices. This information is called a trap. The SNMP Agent
sends the trap to the Management Agent to let the network administrator
know when one or more conditions have been met. Now, some people think
that SNMP only alerts the network administrator when bad things are happening, or when bad things are about to happen. As a matter of fact, these
traps are basically neutral. A trap does not make a judgment; it just notifies
the Manager when something it is supposed to track actually happens. This
includes the following events:
Cold start or warm start During a cold start or warm start of a device,
the Agent will reinitialize its configuration tables.
Linkup or Linkdown This trap is generated by a network interface card
(NIC) when the Agent fails or has to reinitialize to reestablish a connection.
Authentication fails An authentication failure occurring when an
SNMP Agent receives a request for information from an unrecognized
SNMP Manager.

SNMP Managers authenticate with SNMP Agents by the use of something

called a community name. An SNMP Manager can manage a community. An
SNMP Agent can be a member of several communities.

egpNeighborloss The Exterior Gateway Protocol (EGP) is a way for

various networks to communicate with each other. It is a type of routing
protocol. If one of the systems that are supposed to be talking with
another EGP network loses contact, this is an opportunity for an SNMP
egpNeighborloss trap to be sent.
Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

214

Chapter 6

Server Functions

In addition, there are many vendor-specific traps that can be defined.

How Are Traps Defined?
How does an SNMP Agent know what kind of information an SNMP Manager is really interested in? For example, there are dozens of events happening on your network each and every second. What should the Agent trap and
what should the Agent ignore?
As I said, the traps are defined in a database called the Management Information Base (MIB). Depending on the types of traps defined, MIBs come in
various levels, called MIB I and MIB II. These MIBs can be generic, or they can
be defined by a specific vendor for their product. For example, lets say your
network runs a messaging platform from Novell, GroupWise. You have a
GroupWise Internet Agent (GWIA) configured to handle all the inbound and
outbound messages. In that case, Novell GroupWise may have defined a MIB
that tracks the traits of various inbound or outbound messages. It is up to the
administrator to set a baseline and define when these events become important
enough to report. Using the GroupWise example, an administrator may set a
limit so that an alert notification trap is sent to the SNMP Manager if the
GWIA has handled 500 outbound messages in 3 minutes. This may be a sign
that the Internet Agent has been compromised, and your e-mail system is being
used for spam.
Once the condition has been met, the SNMP Agent will form a packet that
lays out the following information:


The SNMP version, usually version 1 (V1) or version 2 (V2).

The Community name of the SNMP Agent.

The type of SNMP trap.

The Enterprise, which the corporation or organization that actually

originated the trap. This shows up as a series of integers like
.1.3.6.1.4.1.x.

The IP address of the SNMP Agent.

The generic trap type, as mentioned above.

A specific trap type.

A time stamp that indicates when the actual event occurred.

An object ID.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Management Protocols

215

The packet is sent to the SNMP Trap Manager through UDP port 162.

Desktop Management Interface (DMI)

The Desktop Management Task Force (DTMF) has taken the idea of
SNMP to the desktop. The DMI specifications lay out a system that is very
similar to the way SNMP operates. Just like SNMP, there is a central collecting point of information and each configured desktop will report to
that central site.

Architecture of DMI
DMI is made up of four pieces: the Service Layer, a Management Information
Format (MIF) Database, a Management Interface (MI), and a Component
Interface (CI). The Service Layer is the piece that acts like the information broker between all the manageable products and the management applications.
The MIF database would be like the MIB database in SNMP; it defines the
standard manageable attributes of the devices installed in the server or in a PC.
The Management Interface allows the DMI applications to actually
gather the information and manage the computers, their components, and all
the peripherals. The Component Interface is what allows the components to
be acknowledged by all the applications that make use of the DMI Service
Layer. It is the CI that gets the calls for the real-time dynamic instrumentation information from the products that can be managed by the DMI. The CI
gives the hardware vendors a known set of values to address in their applications. This allows the vendors to provide more management features and
advanced functionality in their products.

How It Works
At its simplest level, all DMI does is provide two services: It stores information
about various components and then it gives the ability to query the component
and change the information.
The component information is stored in the MIF file. Each DMI-compatible component has its own MIF file, and that file contains information
such as where the equipment is located, the name of the manufacturer, and
when the component was installed. This information is added to the database by the Service Layer when the equipment is installed.
This magical Service Layer is software installed on the desktop computer
that uses DMI. It can be part of the computers operating system, or it can

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

216

Chapter 6

Server Functions

be an add-on. The DMI management application will use the Service Layer
to get information from the desktop component. This information can be
queries as simple as, Who is the manufacturer? or Change the I/O
address to 300 and the interrupt to 3. The DMI management piece can also
accept messages from components, like the message from a printer that notifies the administrator that it is out of paper.
So there are several ways of gathering information about what is happening
inside of a server. Now lets take a look at the face the server gives to the world.
In this case, we are looking at what role the servers play in the network.

Real World Scenario

In the past several chapters, we have looked at how to make the best use of hardware to maintain a high availability server. Make no bones about it, as long as
your server stays "high availability," your workload decreases. Now, I dont know
about you, but one of my goals has always been to decrease my workload!
Another way that you can decrease your workload is to provide yourself
with the tools necessary to be proactive and know what is happening on
your network. That is one of the reasons you can use SNMP. SNMP allows
you to baseline your network during the good times. You can then use this
information to compare the performance during the known good periods
to the performance during times that may not be so good. SNMP is a great
way to keep an eye on your network, and to use the baseline to watch
trends and see if there are things you need to do to prevent problems
before they happen.
DMI is another great tool for keeping you sitting in your seat rather than running all over the company trying to chase down an end user problem. Depending on the implementation, you can take over a users desktop to see what it is
that she is doing, and you can watch the errors that result. You may be able to
answer questions without ever leaving the comfort of your own Cubicle, Sweet
Cubicle.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Server Types

217

Server Types

As part of this objective, we are going to take a look at the way your
server can be used. Gone are the days of the generic, multi-role server: It
seems that every server you bring online has a specific task that it is designed
to do. In this section, we are going to take a look at a bunch of them.

Servers as Gateways
If you were to ask the average end user how the network works, he would kind
of give you a blank look, and if he answered at all, the answer would probably
revolve around smoke and mirrors and Pure Computer Magic (PCM). Well, if
ever there is PCM, it is in the role of a server as a gateway. Most of us have
heard the term gateway at one or another time during our careers, but for
many, what a gateway does remains somewhat mysterious.
Think of a gateway as a translator of sorts. Let me put it to you this way.
Through my life as an instructor, I have met many fascinating and brilliant
people, but people who speak multiple languages still amaze me. You see,
there is an old joke that goes something like this:
What do you call someone who speaks three languages? Trilingual.
What do you call someone who speaks two languages? Bilingual.
What do you call someone who speaks one language? American.
Thats me. I really tried, but French eluded me in grammar and high
school and the only thing I got out of Spanish in college was a crush on the
teacher. I really can only speak English.
Recently my wife and I were invited over to meet the parents of a friend
of ours. Enzo and Colleen are truly multilingual. Enzo speaks Italian (his
native language), English, Spanish, and several others. His wife matches him
language for language with some Thai or Vietnamese thrown in for good
measure. When we went to Enzos house, we met his parents, who only
spoke Italian, Spanish, and something else that didnt make a mark on my
radar screen. Colleen was inside getting dinner, and Enzos dad and I were
on the deck trying to communicate. It wasnt going well until five-year-old
Mateo joined us. You see, Mateo could speak English and Italian, so he
translated. Mateo was my gateway!
Network servers used as gateways come in a variety of forms. For example, earlier, I mentioned a GroupWise Internet Agent. That is a gateway,
because it translates outbound messages from GroupWise to Simple Mail
Transfer Protocol (SMTP). On the e-mail system, there may also be an
Exchange or Notes gateway, which translates messages from Exchange or
Notes to GroupWise.
Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

218

Chapter 6

Server Functions

Gateways basically translate from one form of network communication to

another form of network communication. If you remember your Network+
training, you should remember the seven layers of the Open System Interconnect (OSI) model. These kinds of gateways will operate at the upper levels of
the OSI modelsay, the Application, Presentation, and Session layers.

Keep in mind that whenever you talk about the OSI model, nothing is ever
carved in stone. So if you are thinking to yourself that you know of a gateway
that goes to the Transport layer and below, you are probably right!

Server as a Router
What does a router do? It routes. It routes packets from one network segment
to another network segment. That other network segment could be on the
other side of the room, the other side of the enterprise, or the other side of the
world. Back before I became an instructor, I had a real job as a network administrator. It was really a scary time in my life, because I actually had to work for
a living! Anyway, I managed and administered two servers that provided file and
print services for about 500 users.
One of the servers had enough slots to put five network cards in it. There
were three Ethernet cards, a Token Ring card, and even an ArcNet card. Yes,
we have already established that I am an old guy and have worked with some
really ancient forms of technology. Drop it and lets move on. The point here
is that this single server acted as a router, providing users on the ArcNet segment print services from a printer located on either the Token Ring segment
or any one of the Ethernet segments.
For a server to act as a router, it usually needs at least two network cards
in it. For example, take a look at Figure 6.1.
FIGURE 6.1

Server as a router

Internet
Internal network

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Server Types

219

In this case, you have a server that is acting as a portal between the corporate internal network and the Internet. Any packet exiting the internal network on its way to the outside world has to go through the server, and vice
versa. It is up to the server to direct those packets to the appropriate locations.
It is acting as a router.
A router is an intelligent device. That means that when the router acts on a
packet, it provides that packet with directions. Think of it this way. Assume
that you are at home and need directions to the local shopping center. When
you stop to ask for directions, the woman who provides the directions takes all
sorts of factors into consideration. These can include things like weather, traffic, roads that are closed, and detours; with an intimate knowledge of the area,
she may even provide you with a shortcut. In a similar way, the server acting
as a router can make intelligent decisions about how to get from point A to
point B.
Routing occurs at the Network layer of the OSI model. Another form of
packet management is the server acting as a bridge.

Server as a Bridge
While routing acts at the Network layer of the OSI model, bridging occurs
at the Data Link layer. That means that when a server acts as a bridge and
works on a packet, it is using the media access control (MAC) hard coded
address from the network card. While a router is used to intelligently route
packets from one network to another, a bridge is designed to join network
segments and manage the packet traffic between the segments. Lets look at
it this way.
Assume that you are working for a company that is growing really quickly.
You have a network segment that has about 40 hosts configured to use an Ethernet connection through a hub. You are called into your bosss office and told
that next week the company will be adding another 40 employees, so you are
responsible for connecting these new computers to the network. You know that
having 40 computers on an Ethernet segment will provide acceptable performance, but 80 hosts on the same segment will cause the performance to degrade
to the point of being unacceptable. Somehow there has to be a way to manage
the internal network traffic to ease the load. Enter the bridge.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

220

Chapter 6

Server Functions

FIGURE 6.2

Simple network map with a bridge

Workstation B-1

Workstation A-1

Workstation B-2

Workstation A-2
Segment B

Segment A
Bridge

Workstation B-3

Workstation A-3

Workstation B-4

Workstation A-4

Workstation B-5

Workstation A-5

Take at look at Figure 6.2. Now, I will be the first to admit that this is an
oversimplified explanation, but sometimes those are the best kinds. You will
notice that all the workstations on the left side of the graphic (Segment B) are
numbered B-1 to B-5, and all the workstations on the right side of the
graphic (Segment A) are labeled A-1 to A-5. The goal is to implement a
device that segregates the segments to reduce traffic. So, here is how it works
in an Ethernet implementation.
Suppose that machine A-1 wants to talk with machine B-1. In an Ethernet
implementation, a packet is put on the network segment and all the workstations and the server will receive that packet. Lets assume the server has
just come online and hasnt learned anything about the network. It will just
be hanging out listening and learning. So, the server receives the packet from
Workstation A-1 and makes a note of the workstations MAC address and
the fact that it resides on Segment A. Meanwhile, Workstation B-1 replies to
Workstation A-1. Until B-1 replied, the bridge had no idea where it resided,
but now it gathers in the packet, makes note of B-1s MAC address, and sets
it up for Segment B.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Server Types

221

Now lets assume that Workstation B-5 wants to send a packet to B-1.
When the packet gets to the server, the server does several things. First it makes
note of the MAC address to B-5 and the fact that B-5 resides on Segment B.
Then it looks at the destination address, and sees that the packet is going to
another workstation on the B segment, B-1. At this point, the bridge kills the
packet from crossing over to Segment A. The bridge knows that B-1 is on Segment B, and therefore the workstations on Segment A have no reason to
bother with a message for a workstation on Segment B. By killing the packet,
the bridge is, in effect, lowering the network traffic on Segment A.
Look at Table 6.1. At this point, this is all the bridge knows about.
TABLE 6.1

The Bridge Is Learning

Segment B

Segment A

B-1

A-1

B-5

Now, lets say that Workstation A-3 sends a packet to Workstation A-5.
Since the bridge doesnt know about either of these workstations, it makes
a note of the MAC address for A-3, and figures out that it is on Segment A.
Since the bridge doesnt know where A-5 is, it lets the packet go to both Segment A and Segment B. When Workstation A-5 answers, the bridge then
knows that A-5 is on Segment A. Table 6.2 shows what the bridge knows at
this point.
TABLE 6.2

The Bridge Is Learning Redux

Segment B

Segment A

B-1

A-1

B-5

A-3
A-5

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

222

Chapter 6

Server Functions

As you can see, the missing pieces are becoming filled in. To finish the process, lets say that Workstation A-2 sends a packet to B-3 and A-4 sends a packet
to B-2. The bridge notes where A-2 and A-4 are located, and keeps track of the
MAC address for both workstations. The packets are then sent through to both
segments. Workstations B-3 and B-2 answer their counterparts, and when the
communication is done, the bridge now has the information shown in Table 6.3.
TABLE 6.3

The Bridge Is Still Learning

Segment B

Segment A

B-1

A-1

B-2

A-2

B-3

A-3

B-5

A-4
A-5

The bridge is missing B-4, but it will pick that one up along the way. Now,
lets assume that Workstation A-4 wants to talk with A-1. The bridge knows
that both workstations are on the same segment, so it will not let the packet
get through to the B segment. Therefore, it is limiting and managing traffic.

As I mentioned, bridges operate at the Data Link level and routers operate at the
Network layer. You can use a rule of thumb that says the higher up the OSI model
you go, usually there is more expense involved. The higher up the OSI model you
go, there is usually more done to the packet, so the communication is usually
slower.

With a router, the server was connecting the network to other network
segments and re-addressing packets and directing them to their destination.
With a bridge, the decision is made to let the packet through to another segment or to block it. There is no re-addressing or redirection involved.
Now, lets move a little higher up the server ladder to look at a server that
provides protection to the network.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Server Types

223

Firewall Server
Just like a lot of the other things we have been talking about in this book,
firewalls come in a variety of shapes and sizes. You can buy firewalls to protect entire networks, or you can even buy firewalls to protect workstations
while they are dialed into the Internet. Depending on the amount of money
you want to spend, firewalls can range from free to hundreds of thousands
of dollars. While firewalls vary in their complexity, they do have a lot of
things in common: First and foremost, they are designed for security.
The firewalls we will look at are designed as protection for networks or
for network segments. Take a look at Figure 6.3. This is what is called a
Demilitarized Zone (DMZ) or a screened subnet.

Firewalls can also be used to segregate two segments of private networks. For
example, the research and development department may have a firewall segregating it from the rest of the network.

FIGURE 6.3

Firewalls in a DMZ (screened subnet)

IBM-compatible
Web Server

Mail Server

IBM-compatible
Internet
Firewall

Firewall

IBM-compatible

IBM-compatible

IBM-compatible

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

224

Chapter 6

Server Functions

You will notice that there are two firewalls in this diagram. These two
firewalls create kind of a no-mans-land where servers that need to be publicly accessible can be placed. In this example, there is a Web server and there
is also a mail server in the DMZ. The actual private network is located
behind the second firewall.
How do firewalls work? There are several different kinds but they all
work on the same principle of allow or deny. Here is an example. Using the
graphic shown in Figure 6.3, assume you were located somewhere on the
Internet and wanted to get to that Web server. You knew the URL was
www.server_behind_the-firewall.com, so you entered that in from your
browser and pressed Enter. Your computer would resolve the name
www.server_behind_the-firewall.com to an IP address, and then fire off
a packet directed to that IP address at the port of 80.

Port 80 is what is what is referred to as a well known port for HTTP. Think of
a port this way. Suppose I wanted information about your network and called
the phone number for your company listed in the phone book. As soon as
someone answered, I started asking questions about your network. What kind
of answers would I get? Not good, probably. If I asked for your extension and
started asking questions, I would get the right answers. Ports are like extensions. If I address something to the right port, I will get the information I need.
Well known ports are listed in a file called SERVICES. The location of the file
will vary by operating system.

In order to get to that server, your packet has to go through the firewall.
When the packet arrives at the firewall, the firewall examines it and makes a
decision whether the packet should be allowed or denied. The decision can be
based on a variety of things, including the port address, the protocol, the source
address of the packet, the destination address of the packet, the time of day, the
day of week, or whether you are authenticated into the network or not. In this
example, the packet is handled and the firewall determines that your packet can
be granted access to the URL (or actually, the IP address associated with the
URL) www.server_behind_the-firewall.com at port 80. Your communication goes through and you successfully reach the Web page.

Firewalls can be hardware, software, or both. They can come as a standalone

box like a PIX from Cisco or Checkpoint from Nokia, or it can come as a network
appliance, or just as a software implementation. No matter how it comes, it is
only as good as its configuration.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Server Types

225

If, on the other hand, the port for HyperText Transfer Protocol (HTTP)
had been blocked, your packet would have been discarded. No error messages would have been returned. As far as you were concerned, you would
have received the ever-popular 404 error, Page Not Found.
Why are there two firewalls in the diagram? This is a case of added protection. Earlier in the book, we mentioned things like fail over and redundancy. This is just another case. Assume that some intruder had been
attempting to hack his way into your network. If the hacker made it past the
first wall and gained complete access to the DMZ, he would also have to
defeat the second firewall (or maybe even a third firewall) to get access to the
private network. Is this to say that having two firewalls is the be-all, end-all
in stopping hackers? Not at all. Depending on the sensitivity of your data
and the persistence of the hacker, almost anything can be compromised
eventually. However, you want to make it so difficult for the hacker to get
in that either you will notice the attack and make changes to counteract it,
or the hacker will determine that it is more fun to go play in someone elses
backyard and will leave you alone!
As I mentioned, there are several ways for the firewalls to make these decisions. The first is by the port that is being addressed. Firewalls can also use
an access control list (ACL) where, if you are authenticated into the network,
you can pass through the firewall, and if not authenticated, you do not have
access. There are also firewalls that protect the network by changing protocols. For example, you may have a firewall that switches from IP to IPX.
Another type of firewall is the dynamic state list server. This list resides on
the firewall and is changed as new communications are opened and other
sessions are deleted. When a communication session is opened between two
machines, the communication session is logged and only the initiating computer is allowed to communicate with the recipient. If a hacker were to
attempt to slip a packet into the communication stream, the packet would be
discarded.
Firewalls not only serve to manage inbound packets, they also can track
packets going to the Internet. This is useful in a variety of ways, not the least
of which is to track Internet usage by your users. Several studies have shown
that the largest area of corporate theft revolves around the time lost by
people inappropriately using the Internet. With a firewall, you can specify
what sites or IP addresses people can access or you can deny people access to
specific sites.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

226

Chapter 6

Server Functions

Proxy Server
Time to go look at another figure. Look at Figure 6.4. This is a network diagram with a Proxy Server inserted into the mix. In this case the Proxy Server
is located behind the second firewall, but it doesnt have to be that wayit
can actually be located in the DMZ, or even in the open. It can be located
anywhere a host on the private network can get access to it.
FIGURE 6.4

Server as a Proxy Server

IBM-compatible
Proxy Server

Web Server

Mail Server

IBM-compatible
Internet
Firewall

Firewall

IBM-compatible

IBM-compatible

IBM-compatible

A Proxy Server really has two business reasons for being in existence.
The first goal is to provide faster response to frequently accessed Web
pages. The second goal is to minimize the amount of network traffic generated between the private network and the Web.
It works like this. The user Brandice, on the private network, wants to
access the Web page at www.myperfectwebpage.com. Brandices browser
has been configured so that it points, not directly to the Internet, but to the
Proxy Server instead. In this case the browser asks the Proxy Server if it has
the Web page, www.myperfectwebpage.com. The Proxy Server doesnt have

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Server Types

227

it at present, so the Proxy Server goes out and gets the Web page, caches it,
and returns the results back to Brandices computer. A short time later both
CJ and Courtney send requests to get www.myperfectwebpage.com, and
just like before, their computers go to the Proxy Server and ask for the Web
page. In this instance, the Proxy Server already has the Web page stored in
cache, so it is immediately returned to their computers, without anyone having to cross the network and download it.
A proxy server can perform many firewall-like functions. For example, it
can be configured to allow or deny access to certain Web sites. Many Proxy
Servers also have the ability to provide network address translation (NAT)
to the internal network. NAT is a function where the Proxy Server or firewall
can have a public IP address, while everything on the private side of the network can have an IP address from a different, private range. This helps conserve registered IP addresses and makes it more difficult for hackers to access
systems behind the Proxy Server.
Another business use for the Proxy Server is to configure it to be a Reverse
Proxy Server. In Figure 6.5, you can see that the Proxy Servers sit between
the corporate Web server and the Internet. These servers are configured to
cache the information from the Web server, and provide the content when
queried from the Internet.
FIGURE 6.5

Reverse Proxy Server

Proxy Server

Internet
Web Server

Proxy Server

Proxy Server

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

228

Chapter 6

Server Functions

By using a Reverse Proxy Server, people accessing the Web site can be
directed to several different addresses, providing for quicker response. Look at
it this way. Suppose that you were getting ready to start a world-class e-business. You were going to use the Amazon.com business model of selling tons of
stuff over the Internet. You want to make sure that people have rapid access
to your Web site, no matter where they are located. So, you configure your
infrastructure so that you have your primary corporate Web server in your
home office located in Minneapolis. You then have Proxy Servers set up in
New York, London, Paris, Moscow, Delhi, Jakarta, Sydney, Hilo, and Los
Angeles. This way, the Proxy Servers point back to the main Web server, and
any change you make to the corporate Web server in Minneapolis will be replicated out to the Proxy Servers almost immediately, saving on the cost of
maintaining all those additional Web servers. Reverse proxy also provides
some redundancy. If you have only one Web server, and it goes down, your
Internet presence is now hosed. If, however, you are using Proxy Servers and
one of those go down, the others can pick up the slack.

Hosed is a high-level technical term. Actually, it is an acronym standing for

Hardware Or Software Error Detected. Just thought you would like to know!

The next two types of servers usually come in one package because it
makes it much easier, and usually more secure, to access a database using a
client/server connection. Lets start by looking at the database server.

Database Server
Here is another one of those catch phrases of the new millennium, data warehousing. The term information age is another one of those terms that has been
beaten to death but, even though it is trite, it is still true. Businesses succeed or
fail based on decisions that the management team makes, and the management
team can only make good decisions based on the information they have to
work with. That information is stored on one of the companys servers, and
when someone has a question, the numbers need to be crunched to come up
with the appropriate answer. Lets take a look at an example.
Usually when you start talking about a database server, you are talking a
medium to large organization. But the company we are going to look at is a
small company with less than 10 employees. My wife, Bobbi, started the
company back in the mid-90s and has successfully steered the company
through good times and bad. Her company, For Your Instructors, is a brokerage service for independent computer trainers just like me.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Server Types

229

Now, most trainers I know are really good geeks. Some people I know have
gotten offended when I called them a geek, but my friends understand that the
term is a term of endearment. My best friends are geeks. I have stopped listing
certifications on my business cardsnow it will just say geek. Being a geek
means more than just having a clean pocket protector. Most geeks are very
comfortable with computers and very happy to spend the better part of their
day relating to either computers or the people who relate to computers. We
usually have a somewhat difficult time relating to real people, because they
dont understand the mysteries of what we deal with every day.
Enter the trainer. These people are usually high-level geeks. Especially the
ones who go out on their own and try to make a living all by themselves.
They have the geek part down to a science, but now they have to sell themselves. Selling requires people skills, and many of us are great with computersit is people we have trouble with. My wife, on the other hand, is lousy
with computers but great with people. It is okay that I say that, because she
will never read this far into the book. She would have been bored to tears and
asleep by the time she hit the chapter on RAID.
She is so good with people, as a matter of fact, that she can even operate
in the geek world and have people relate to her. This is a match made in
heaven. Bobbi offers an invaluable service to many of the finest trainers in
the world. She relieves them of the responsibility of dealing with people, and
gives them the opportunity to work their trade to the best of their ability.
She, and her very efficient staff, match trainers with training centers and
classes, and take a commission for doing all the work.
Now, that means she has to have some way to keep track of tons of data.
We are talking a mess of information. For example, the database that her
company uses has information on thousands of trainers and thousands of
training centers. She has information on what classes the trainers can teach;
where they fly out of; and what their favorite airline, rental car company,
and hotels are. On the training center side, she tracks who is the contact person, what they have used in the past, what they have need of in the future,
what they have been billed, and what they still owe. All that information is
contained on a database server that is the lifeblood of that company. We
make darn sure that the database is backed up regularly and taken care of
like a newborn baby.
By accessing that database, she can do a search to determine which of her
trainers are certified to teach what class. She can determine if they are open
the week of the class and can also find out all the other information she needs
to know. She has a server that is dedicated to nothing other than running and
caring for her database applications.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

230

Chapter 6

Server Functions

Special care should be taken when backing up database servers. Databases

are often open for long periods of time, and some backup programs will not
back up open databases without a special add-on piece. Sometimes these
pieces are free, and sometime they cost, but they are always vital. If you have
a database server, make sure that it is backed up, and make sure it is backed
up using a program that will handle open files.

Because database servers are going to be crunching tons of numbers and

also handling massive amounts of indexes and searches, the server that
houses this vital function should be very long on memory, processor power,
and disk space. Redundancy and backups are exceptionally important. On a
scale of 1 to 10 on the importance meter, the database server is way high up
there. Most companies can operate without e-mail, many companies can get
by without some applications, but if the data is not there, everything stops!
One way of protecting that data is to use client/server applications.

Client/Server
A friend of mine called recently and asked me to try to explain to his wife what
client/server computing was all about. The thing that made this conversation
interesting was the fact that he has a masters degree in Information Technology.
He knew what client/server computing did, and why it was good, but he really
wasnt 100% sure how it worked.
When you think of client/server computing, think of going to a very fine
restaurant. When you walk through the front door and give them your name
and time of reservation, you become the client, and the staff becomes the
servers. When you order, you ask the server for a meal, and the server takes
the information back into the kitchen, and a short time later, you are basking
in the glow of a gourmet meal. Meanwhile, the server is available for your
every want and need. Take a sip of water, and they are there to fill your glass.
Finish your wine, and they are there to offer you more. For the entire time
you are in the restaurant, you are being waited on hand and foot.
Is the waiter or the chef or the maitre d doing anything that you couldnt
do for yourself? Absolutely not! You can cook, pour your own water, and
open your own wine, but it is special when it is done for you. Besides, I dont
know about you, but no matter what I order in a restaurant, it always seems
to taste better than it would if I ended up cooking it myself. So, how does this
relate to computing?

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Server Types

231

Lets take a look at a common implementation of client/server computing,

e-mail. Now, many people think that e-mail systems are smoke and mirrors,
and they have no idea what happens in the background to make electronic
messages magically appear in their In box. The truth be told, e-mail is just a
very simple database.
So, lets look at what happens when someone sends an e-mail message. Lets
go back to our friend in the restaurant. That meal was so very good, our friend
did what any other person would doswiped a menu! When he finally got
back to work, he made inappropriate use of a corporate resource and scanned
the menu into a very large 8MB file. When the scan was done, he addressed an
e-mail message to 250 of his closest friends, attached the menu, and then sent
off the message. Now, in the background, the e-mail software on his computer
(think Outlook, or Notes, or GroupWise client) gets access to a database and
starts copying this large attachment into the database. Now, I dont know
about your users, but every user that I have ever been associated with does not
have patience as a middle name. So, right in the middle of the file copy into the
database, the end user shuts his computer off. What happens? The database
becomes corrupted. Now, if the database only served the user who shut his
machine off, he would have gotten what he deserved. In the case of e-mail, on
the other hand, these databases are usually shared among large groups of
users, all of whom will now suffer because this person ran out of patience. We
can call this scenario direct access to the database.
Contrast that with client/server access. Look at Figure 6.6. Notice that
there are three pieces to this puzzle. The first is the e-mail client, which is running on the workstation. The next is the server process, running usually (but
not always) on the same server on which the mail databases are stored.
Finally there is the database server, where the e-mail databases are stored. In
this case, anytime the client wants to access the databases, it goes through the
server process. This means the client asks the server process for something,
and the server process then returns what it asks for. The client never has to
actually access the database, which means than an impatient user cannot
corrupt the database.
FIGURE 6.6

Client/server computing

E-mail client

Database server
Server process

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

232

Chapter 6

Server Functions

Lets go back to the scenario above. Assume we have switched over to a

client/server method of accessing the database. Same 8MB file, same impatient
user, same 250-user mailing list. In this case, the client tells the server process
(usually using TCP/IP) that it has a job for it. The server answers and starts
accepting information from the client. Halfway through, same exact thing happens: The user gets impatient and shuts his computer off. When it happens this
time, though, the server process is in charge. It stops getting information from
the client so it backs the transaction out of the database and cleans up neatly
after itself. No corrupt database for all the other users to suffer through.
The client/server architecture is used in most types of database applications.
The client talks with a server process; the server process accesses the database
and returns the information to the client. The user never has his fingers in the
database, so he cannot actually go in and goof anything up.

Application Server
Just like its name suggests, the purpose of an application server is to serve up
applications. This can best be described as the economy of scale.
One of the most daunting tasks that a network administrator has to handle is the management of applications. They have to be installed, updated,
patched, tweaked, prodded, and poked to get the best possible performance
from them. As soon as you have an application optimized for your environment, another patch comes along, and you have to start over.
The problem with application management becomes apparent when you
think about the two ways there are to handle the applications that your users
access every day. The first way is to have the application reside on the workstation. This means that someone has to go around to gosh knows how many
workstations and install and configure the application. Even if you use a program that will push the application installation out to the hard drive of your
users, there is still the work of testing and retesting the rollout, and then troubleshooting the whole thing when someone deletes the wrong directory or a
user corrupts a file.
With an application server, there is one installation. The users are then
granted access to the server and to the application though a series of rights or
permissions. There is only one place that has to be maintained. If there are files
that need to be accessed locally, there are ways of pushing those files down to the
local workstation from the server without end-user intervention. In some cases,
there are ways of having the application check to make sure none of the files on
the local workstation have been corrupted. If corruption or deletion has
occurred, the application basically reinstalls itself on the local workstation.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Server Types

233

As complex as applications are these days, it is no wonder that the companies

that write them need to fix them occasionally. Call it a patch, fix, service release,
or whatever euphemism you want, something didnt work right, and now there
is a program that can fix it. Having the application in only one place (or a few
places in an enterprise network) makes this task a lot less daunting. Instead of
visiting hundreds or thousands of workstations, the task can be done once and
be done with it.
On the minus side of having an application server, there is the problem of
single point of failure. Lets look at an example. In most companies, the application that is probably used the most is the word processing application. Now,
it doesnt matter what brand or version of word processing application you
use, the fact is, much of your company depends on this application to carry out
their tasks for the day. If you have an application server in place, all these users
will be accessing the word processing application from the server. If that server
is unavailable for whatever reason, you have a whole bunch of people sitting
around on their thumbs, earning a salary for doing little or no work. This is
usually not considered to be career enhancing for the person in charge of the
application server. Another high visibility server is the box that brings in and
stores e-mail.
There are other types of application servers as well. For example, a Web
server can be an application server. Another type of application server would
be a Citrix or Microsoft Terminal Server implementation, where a network
server was really being used in the manner of an old mainframe. With Citrix
or with Terminal Server, all the processing is being done at the server, so it
is actually running the application.
The thing to remember about application servers is that they serve up a
type of application for the client or for the user.

Mail Server
When I teach messaging classes, which I do quite often, I tell my students that
electronic messaging in any form can be broken down into three pieces: process,
item, and location.
Process is always an executable. It can be an exe file, a com file, a dll file,
an NLM, or whatever Unix uses, it doesnt matter. A process is always an
executable that has to run on some form of a server.
Item is the message that is being moved. This message can be an e-mail,
a voice mail, a pager messageit doesnt matter, but it is always some
form of message.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

234

Chapter 6

Server Functions

Location is where the message is currently being stored and the location
where it will be stored when it reaches its next destination.
As you can well imagine, a mail server seems to serve two-thirds of the
equation very well. The mail server will usually be running the process that
will either deliver the mail or manage the mail system, or both.
Given the visibility of e-mail in todays business marketplace, this is
another of those high visibility type servers. People tend to get really cranky
when they cant get their e-mail, or when their e-mail does not arrive in a
timely manner. The e-mail server is one of those boxes that should get special
treatment!
If you are lucky enough to be charged with preparing a list of specifications
for an e-mail server, think about what it does before rushing out to buy one. First
of all, there is going to be a lot of ongoing communications with the server, so
the network communications should be top-of-the-line, especially if this server
is going to need to contact the Internet, or another mail server somewhere else
on the WAN.
Secondly, there is going to be a high number of very small disk reads and
writes as all these small entries get made into the database. Remember, in the
world of e-mail, it is not only the large mail message that is written to the
database, it is also the smaller piece that says the message has been opened
or delivered. This is called the status message. Since most of these are rapid,
small read/writes, you should make sure that your disk subsystem is as fast
as you can afford. Finally, most mail server processes will use some form of
cacheing, whether it is at the server side of the equation or the mail process
end. That means providing as large amounts of memory as possible.
One of the other questions that I constantly get asked is, How much disk
space, per user, should I allocate? There is no right or wrong answer to that
question. The only thing I can tell you is that no matter how much you allocate, it will not be enough. It is a physical law that nature abhors a vacuum
and there is no stronger vacuum than empty disk space. If your e-mail system
is not hooked to the Internet, you may get by with 25MB or 30MB of storage
per person. Once you make that Internet connection, however, all bets are
off. For grins and giggles, I just checked how much disk space my remote
mailbox took up on my laptop, and it is up to about 125MBand my mailbox is not full. I routinely go out and empty my trash and clean out the sent
items folder. I will bet there are not a lot of your users that will do that. So,
you can always use more disk space.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Server Types

235

Another thing to keep in mind about e-mail systems is that they are somewhat fragile. Because they are databases with the potential for access by so
many people, they can be corrupted easily. Anything you can do to protect
the mail server would help you later.
E-mail servers that are connected to the Internet will use the Simple Mail
Transfer Protocol (SMTP)/Multi-Purpose Internet Mail Extensions
(MIME).
I would hazard a guess that most e-mail systems in place today have some
sort of inbound and outbound access to the Internet. As you design your email system and the placement of your mail server, be sure that it is secure.
You may want to place it behind the firewall and open up the appropriate
port addresses. If you have people who are going to download files from
your company, you will have to do the same type of thing with the File
Transfer Protocol (FTP) server.

FTP Server
Depending on the type of company you work for, you may have to provide
access to information in a file format. This information may be data, it could
be a patch, or even the delivery of an application. Somehow, you have to create a way for users to attach to a server and either copy the file down to their
workstation or take information and transfer it up to the storage server.
This book is a prime example of how this process works. What you are
holding in your hands is the result of months of work by a very large team of
people. I am just a very small cog in a very large wheel. Everything, though,
does start with me. Once I have finished writing a chapter, I have to transfer
that chapter and all the art to my developmental editor, Elizabeth Hurley, or
to my production editor, Judith Hibbard, or to the editor who really writes the
book, Linda Stephenson. In addition, there are also the art people and the technical editor, who all need to access the raw chapter and begin to do whatever
it is that they do. There must be a central repository for the information, and
that is an FTP server that Sybex hosts. When I get done with a chapter, I zip
all the parts together and then send it off to Judith. She looks it over, and posts
it on the FTP server, so everyone can download what they need and tweak,
twist, torment, and do whatever to it before putting it back.
FTP is a standard part of the TCP/IP protocol suite. If you are configuring
an FTP server, make sure to have plenty of disk space for storage and plenty
of bandwidth for data transmission.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

236

Chapter 6

Server Functions

FTP normally is not considered the most secure of protocols. You can create an FTP server that can limit access to a certain set of people that can log
in using an e-mail address and password. Other FTP servers may allow for
anonymous access, where anyone can attach to the server and download or
upload information.

SNA Server
When you see the term System Network Architecture (SNA), think IBM and
think mainframe. When you see the term SNA Server, think of a server
designed to bridge the gap between your PC-based, server-based network
and your mainframe. The SNA server product allows users attached to the
PC/server-based network to access the mainframe in a familiar way. Does
this sound somewhat familiar from what we were talking about when we
discussed servers as a gateway? It should, because that is similar to what is
going on here, but in this case, it is doing it with an attitude.
The latest SNA Server product from Microsoft is called Host Integration
Server 2000. It replaces the SNA Server that was part of the Microsoft Back
Office suite of products. This product is designed to provide a portal to the
mainframe that can handle up to 30,000 simultaneous host sessions per
server. Figure 6.7 may give you an idea of how the SNA Server is positioned.
FIGURE 6.7

SNA Server

IBM AS/400

Server-based network

IBM 37XX
SNA Server

IBM 3174

IBM AS/400

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Server Types

237

As you can see from Figure 6.7, the SNA Server acts as a front end for the
mainframes. There are clients available that are Microsoft ActiveX-enabled,
or they can be accessed from the Web. These clients can provide access to
3270 or 5250 systems, as well as the AS/400.
One of the issues that always seemed to be around whenever there were
host connectivity issues was the problem of passwords. A user would need a
password to log on to the server-based network, and then as soon as the user
attempted to access information or resources off of the mainframe, they were
prompted for another password. The new SNA server products can map user
IDs and passwords from the server-based networks to the mainframes to
eliminate this issue.
SNA Servers can also provide virtual private networking services directly
from a client through to the mainframe. This solution allows for users that
are roaming the country using their laptops to access the mainframe securely
wherever they happen to be. Remote communications is always an issue,
which is why your network may include a Remote Access Services (RAS)
Server.

RAS Server
When I think of a RAS Server, I think of a modem farm. The whole point of
a RAS Server is to provide remote access into the network via a Plain Old
Telephone System (POTS).
When Microsofts RAS was first introduced, dialup networking was not
only ugly, it was really ugly. There were a few good solutions, but none that
really worked first time every time. In some cases the dialup solution would
allow the user to connect to various parts of the network, like to a mail
server, but would not allow them to connect directly to the network. There
were products that would allow you to dial in to a system that was connected
to the network, and then take over control of that system to log on to the network. But that was kind of a clunky solution that added yet another level of
hardware between the remote user and the network.
RAS changed all that. With RAS, you can have a server with multiple
modems and incoming lines. Your remote users can dial in to the RAS server
and be logged directly onto the network, as if they were sitting at their desks.
Now, the access speed is not as fast as it would be at the end of the 100MB
Ethernet connection, but that is to be expected.
If you are configuring a RAS server, there are several things to keep in mind.
It is going to be the portal for multiple connections, so it should offer the fastest
response time possible. That means fast, dependable modems in the modem
pool, high-speed dependable network connectivity, and lots of memory.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

238

Chapter 6

Server Functions

If ever you were going to need remote management software, this is it.
You will need it to remotely monitor the server to see who is using what, and
you will also need it when someone forgets how to log off properly. Many
times a modem may hang, and need to be reset. This is a simple procedure,
and it is even simpler if you can remotely manage the server.

File and Print Server

This is truly the legacy server application. File and print services have been
around since the days of NetWare 2 and before. A file and print server provides secure, controlled access to files, as well as management for a variety
of types of printers located all over the network.
File and print services are the core of any network. Of all of the servers
that we have talked about in this chapter, this is probably the one that you
will have the most of. These are the servers where the users home directories
are stored. When the information that is stored in one of those universally
shared directories needs a place to go, this is it. All that information has to
be secured, so that not just anyone can know that Cris has a copy of a file
called RESUME.DOC stored on the network. There are other places that you
dont want just every Tom, Dick, and Mary poking around, like the spreadsheet that contains corporate salaries or the information on the CEOs stock
options. All this information needs a place to live, and it needs a centralized
management location. That is why there is a file server.
Printing has become the water cooler of the new millennium. It used to be
that people everywhere hung around the water cooler to keep up on all the
latest news of what was happening in the company. Now it is the printer.
The file and print server has a big job. It has to take all the print jobs that
come in from potentially thousands of users and redirect those jobs to the
appropriate printer, in the right order. Depending on the type of file and
print services you are using, it might also be up to the print server to make
sure that the client requesting the print job has the right print driver also.
When configuring file and print servers, make sure you have plenty of disk
space and plenty of memory. Have you heard that somewhere before?

Fax Server
Are there any accountants reading this book? If you are, you will probably
have to excuse me here for a few minutes while I pick on you! Bean counters
everywhere hate the inappropriate use of a corporate resource. What is the
most valuable of corporate resources? That would be the employees time.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Server Types

239

Think about what happens when you send a fax from a conventional fax
machine. First of all, you have to print the document out, and walk over to
the printer to get it. You get up from your desk and pass the cube of your best
friend, so you shoot the breeze for a minute. Then you get to the printer, and
there is the new person from Legal, so you spend a minute building a rapport
and welcoming them to the company. When it comes time to grab your print
job, you grab a large stack of paper, and begin rifling through looking for
your particular print job. On the way through the stack you see the agenda
for the next Managers meeting, a memo going out on how to keep down the
high cost of executive lunches, and several sheets printed out with the latest
joke of the day. Then you finally find your document, and it is off to the fax
machine.
When you get there, there is someone already faxing something, so you
spend some time finding out about her sons latest soccer game, and she finds
out about your daughters latest basketball game. Now that we have the
kids update out of the way, your partner leaves and you are left to your
devices. You start the fax, and while you are standing there watching the 10
pieces of paper, with a cover sheet, go through the fax machine, you pick up
the stack of paper in the tray just to see if there is anything there for you. So
you rifle through the stack, and you find out that someone in accounting is
making a bid on a new house, that someone in the IT department is leasing
a new computer, and can you imagine the price the company is getting on the
250 new laptops they are ordering? Once the fax is finished, it is back to your
desk, with a stop at the coffee machine and some conversation with someone
else from payroll. Total time to send the fax: 30 minutes. Total time actually
faxing: 10 minutes. Total time wasted: 20 minutes.
The average user may cost your company $50 per hour after benefits and
vacations, so that fax just wasted about $16 of the companys money. That
is not all that much, but when you multiply it by 50 times a day, 5 days a
week, 52 weeks a year, it adds up.
A fax server alleviates that problem. It provides a central point for faxing
services. In other words, if I am sitting at my desk, and I have a 40-page proposal to fax to a client, all I have to do is print that proposal to the fax server.
The fax server then takes the document, renders it (prepares it for faxing),
dials the appropriate number, and when the phone is answered, it transmits
the fax. Meanwhile, on the other side, when someone faxes something
inbound, the fax server answers, accepts the fax, and routes it to the e-mail
box of the recipient. Everything is handled directly from the users desk, and
the user never has to leave the comfort of her chair.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

240

Chapter 6

Server Functions

If you are configuring a fax server and you will be using a standard phone
line, the inbound and outbound line should be dedicated to the fax server.
The server will be doing large amounts of graphics rendering, which will
require lots of memory and lots of disk space, as well as lots of processing
power. Depending on the number of faxes that you expect the server to handle, you may want to use something like a PCI plug-in card that contains
multiple modems on the same card. Some expansion cards can manage 32 or
more modems. In this case you can have several outbound lines and several
incoming lines to alleviate the expected wait for the fax to be sent. As always,
before purchasing the server, be sure to read the fine manual (RTFM) to find
out minimum hardware and software requirements.
So far, we have been talking about servers that provide a rather visible service to your users, one that they can recognize and understand. Next, we are
going to look at the unsung heroes of the network, those types of servers that
just sit there and do lots of work, although no one may even know they are
around. Lets start by looking at a Domain Name Service (DNS) Server.

DNS Server
By this time in your career, you should have the DNS basics down to a tee.
You should know that it is a hierarchical distributed naming convention that
is used to resolve user-friendly names to IP addresses. The top level of the
DNS tree is the root, which is designated with a . (period). After that, there
are zones that represent the usual divisions that you are familiar with, things
like .com, .edu, .org, and .gov.
Now, you may be wondering how a DNS Server can help in your network.
Well, if you are even thinking about using Windows 2000, be sure to learn
everything there is to learn about DNS, because it is one of the foundations of
a Windows 2000 network. Even if you are not considering Windows 2000,
having a DNS Server on-site can prove to be a benefit to your network, now
and in the future.
DNS, as we have just discussed, is used to resolve user-friendly host
names, like www.myperfectwebsite.com, to an IP address, like
192.168.10.5. You can set up a DNS Server to manage your own DNS
namespace, or you can set up a DNS Server or servers to provide quicker
response to name queries.
When a business is small and first starting out, it connects to the Internet
through an Internet Service Provider (ISP), and utilizes the services the ISP
provides. This can include things like DNS lookup services. Figure 6.8 will
give you an idea of the layout at this stage.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Server Types

FIGURE 6.8

241

DNS Server hosted by third party

Internet
Requestor

DNS Server

In this diagram, the host on the local network puts a request in for
www.myperfectwebsite.com. The local host tries to resolve that name to an
IP address by checking memory, looking at several local text files, and finally
coming to the realization that it has absolutely no idea where that Web site
resides. So it better ask for help. It checks its TCP/IP configuration and finds
that the DNS Server is located at 214.177.158.16. The host then has to
determine if that address is on the local subnet or remote. It is remote, so the
packet gets sent to a router. The router than directs the packet across the
Internet until it finally locates the appropriate IP address. At that point, the
DNS Server checks its information to see if it really knows about the Web
site. Lets assume it does. If it does, it will return the IP address back to the
host. There was a lot of communication going on there that required time.
Look at Figure 6.9.
FIGURE 6.9

Local DNS Server

Internet
Requestor

DNS Server

In this case the DNS Server is moved to the local network. Now, when the
requestor is looking for the name, it will send a packet to the local DNS
Server, removing the router and removing the Internet connection from the
equation. The DNS Server can then resolve the IP address if it knows it. The
response time should be noticeably quicker, meaning you wont get as many
complaints!

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

242

Chapter 6

Server Functions

Using DNS to Manage Namespace

The question you may ask here is how can I use this to the best advantage on
my network? As I mentioned, if you are using Windows 2000, this question
should have already been answered during the planning stages of your network installation because it is so key to the network infrastructure. If not,
lets look at ways that you can use DNS to resolve names internally, say, on
your intranet.
DNS Zones
When you register your domain name, you are, in effect, creating what is called
a zone in the domain namespace. For example, I have several DNS names registered. One of them is technotriage.com. When I registered that name, a record
was created in the DNS registry. Now, after it is registered, I want to manage the
namespace.
I start by creating a DNS zone. A zone is nothing more than a part of the
namespace database that is defined by the resource records contained in the zone
database file. So, if I create a zone, say mn.technotriage.com, all the resource
records that I store in the database will define the zone.
There are several kinds of zones. For example, you can create a primary
zone (technotriage.com) or a secondary zone, which is usually a smaller
chunk of the primary zone (mn.technotriage.com). There will be a DNS
Server configured for each of the zones, and to make sure that they are both
up to date, a process called zone transfer is configured.
Zone transfer sounds exotic, but it is really not. It is simply an exchange
of information, or copying stuff from one spot to another. With DNS zone
transfer, there are two kinds: incremental zone transfer and full zone transfer. With a full zone transfer, all records are exchanged between DNS Servers. With an incremental zone transfer, only the changes that have been made
to records are transferred.

DDNS, DHCP, and IPv6

After you read the sentence above, you may have asked yourself, But how
many changes are there in a local DNS database? Under the old (IPv4)
version of DNS, all changes were manually entered, and so the number of
changes implemented on a network may not have been that great. But, with
the beginnings of implementation of IPv6, more and more operating systems and services are relying on something called Dynamic DNS (DDNS).
Here is how it works. There is a DNS Server capable of handling dynamic

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Server Types

243

updates on a network. Meanwhile, in another section of the network, a

host comes online and immediately requests an IP address from a Dynamic
Host Configuration Protocol (DHCP) Server. The DHCP Server, in a
method that will be described in a few pages, assigns the host all the IP
information it needs to function, including an IP address, default subnet
mask, locations of DNS Servers, IP address of the gateway, and the IP
address of the WINS Server if necessary. All the DHCP Server wants in
return is the DNS name of the workstation. So, the workstation comes
back and provides the DNS name, say gary.psconsulting.com. DHCP then
sends the DNS name and the IP address it just assigned to that workstation
to DNS for automatic update. DNS enters the information in the database,
and zone transfers it out to all the other DNS Servers in the network.
That is why Windows 2000 really, really wants you to use its version of
DHCP and DNS, because it knows how things work. This is a pretty far cry
from the days when Windows Internet Name Service (WINS) was used to
resolve NetBIOS names to IP addresses.

WINS Server
As I am sure you remember from your Network+ class, versions of the Windows network operating system prior to Windows 2000 depended heavily
on a protocol called NetBIOS. As a matter of fact, whenever you installed a
version of Windows, you had to provide the computer with a name that was
unique to the network. This was referred to as the NetBIOS name, and this
is the name that other Windows workstations used to find your computer. In
order to locate that name, though, the workstations all had to register their
name with the WINS Server and also request information from the WINS
Server.
NetBIOS names could only be 16 characters in length, and they had to be
unique on the network. When the computer boots, it looks to its configuration
and finds the IP address of the WINS Server, and then it sends a packet that
will dynamically register the computer with the WINS Server. Many of the
commands used in Windows versions prior to Windows 2000 use these NetBIOS names in order to find and access services.
These unique names can have only one IP address associated with them.
NetBIOS is the protocol that has to establish the computer names, and establish and maintain sessions between two computers.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

244

Chapter 6

Server Functions

NetBIOS is, by design, a small protocol designed primarily for a workgroup environment. One of its biggest drawbacks is its inability to be routed
between network segments. This property is referred to as a non-routable
protocol. By establishing a WINS Service, the workstations that come on line
will not have to use a broadcast method of determining the location of their
neighbors.
WINS is used on Windows NT networks that utilize TCP/IP as the transport protocol.
Another of the protocols and services found on a TCP/IP server is a
Dynamic Host Configuration Protocol (DHCP) Server. This one can really
save you a lot of work!

DHCP Server
Back when the Internet was young, and back when there were very few hosts
attached to it, configuring a static TCP/IP address was not a big thing.
Because there werent many computers hooked to the Internet, every computer had its own unique IP address, and that computer was configured once
with an address and it pretty much stayed the same. Then the Internet
became really popular, and suddenly hundreds of machines needed to be
configured with the TCP/IP protocol suite and receive IP addresses, the
address of the gateway, the location of the DNS Servers, the address of
WINS Servers, and bunches of other information. To make matters worse,
many of these computers were not stable, dependable, stay-in-one-place
desktops. They were laptops (called, affectionately I might add, luggables at
the time) that kept moving around from place to place and still wanted an IP
configuration. To solve this problem, administrators looked to DHCP.
A DHCP Server is in charge of passing out IP configuration information
to any host that asks for an IP address. Once the IP address is given out, it
is up to the DHCP Server to manage that address and make sure that it is
really being used. If the address is not being used, then it is reclaimed by the
DHCP Server and passed out to some other host on the network.
In order to make use of a DHCP Server, you start by configuring the
service. The first thing you have to come up with is some minimal information, like a scope or list of IP addresses the server can assign, the
address of the default gateway or router, the address of a couple of DNS
Servers, and the address of a WINS Server if necessary. Now, the IP
addresses all have to be from the same subnet, and the default gateway is
the gateway that serves that subnet. You will see why that is important in
a minute.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Server Types

245

The services that a DHCP Server provides are pretty much operating system independent. That means if you have a Unix system that is configured to
be a DHCP Server, it can provide addresses to Windows NT machines, Windows 2000 machines, Windows ME machines, Linux boxesanything that
asks can get an IP address with all the other information it needs.
Now you will notice that in that last paragraph I said DHCP Servers were
pretty much operating system independent. I put that phrase in there for
two reasons. The first was to drive my editors nuts because they hate things
like that, and the second reason was to make a point. Remember up above
where we talked about Dynamic DNS and I said that it was up to the DHCP
Server to return the proper information back to the DNS Server for registration? Well, if the version of DHCP Server cant work and play well with the
version of the DNS Server, this whole process will fail. If you are using
Dynamic DNS (like in a Windows 2000 network), it is best if the platform
matches. That doesnt mean that DHCP cannot pass out IP addresses to a
variety of different operating systems, because it can. It just means that DNS
and DHCP have to be compatible.

DHCP Servers have to have a static IP address. A DHCP Server cannot also be
a DHCP Client.

You will note I also said that the IP addresses have to be valid for the same
subnet. That is because of the way the hosts request the services of a DHCP
service. That whole process starts when the host boots up and starts to come
online. The host looks around for configuration information. It looks to the
way it is supposed to be configured and finds that it is supposed to receive all
the information it needs to communicate using IP from a DHCP Server. At
that point, the host sends out a broadcast message looking for an IP address.
A broadcast message means that every host on the subnet will handle the
message. Most of the hosts will ignore the message. The hosts that dont will
be the DHCP Servers. The DHCP Server looks at the request and sends back
a packet to the requesting host with a DHCP Offer. This offer is just that; it
is a packet that says, in effect, Hey, I can give you the information if you
want it. Here is my address, just let me know if you want to take me up on
this deal. The host will respond positively to the first DHCP Offer it
receives, and it will send a message back to the DHCP Server asking for the
information. The DHCP Server responds with all of the information that it
has been configured to give.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

246

Chapter 6

Server Functions

Remember back a few paragraphs when we said that the IP addresses

were subnet dependent? Well, because you usually pass out the address of
the default gateway, and the default gateway has to be on the same subnet,
this makes sense. Also, DHCP requests are broadcast messages. Broadcast
messages are never routed across subnet boundaries. That would be just too
much unwanted network traffic.

The actual DHCP server does not have to be on the same subnet as the requesting host. If it is on a different subnet, there has to be some advanced configuration done before things will work. Either the router or another computer on
the subnet without the DHCP Server has to be configured as something called
a DHCP Relay Agent. The DHCP Relay Agent is simply a host on the subnet that
is configured to listen for DHCP requests. If it gets one, it sends a directed packet
to the DHCP Server, the Server responds to the Relay Agent, and the Relay
Agent responds to the requesting host. It is just like a middleman.

DHCP Servers do not give out IP address information. The information is

leased. Just like the lease that you sign on your apartment, these leases are for
a set period of time, usually around three days by default. Halfway through
the lease period, the DHCP client will send a message to the DHCP Server to
renew the lease. Just like any good landlord, the DHCP Server wants to lease
out its addresses, so it will approve the request. If the DHCP client is brought
down for any reason, the address is returned to the DHCP address pool and
given out to another client on the subnet.
The Internet sure has made life interesting for network server administrators,
hasnt it? All so that your users can communicate using TCP/IP and get out to the
Internet to look at Web pages. But what manages the Web pages? That is known
as a Web server.

Web Server
The workhorse of the Internet, intranet, or extranet is the Web server. The
Web server is a server that runs software that lets it respond to and manage
HyperText Transfer Protocol (HTTP) requests. These servers have got to be
able to handle hundreds of simultaneous requests for information, coming
from all over the world. They have to be able to return Web pages of various
sizes, using various kinds of protocols, quickly, accurately, and securely.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Server Types

247

The term Web server can refer to the actual hardware box, or to the software that carries out the service. If you are referring to the software, Web
servers are varied. There is the Internet Information Service (IIS) that comes
with Microsoft products, there is the Netscape Enterprise Web Server,
Apache Web Server, and even Macintoshes can run a Web server. Some are
available for download and some are available for purchase. Each will run
on a specific network operating system: IIS for instance runs on Windows
server products.
When you enter www.myperfectwebsite.com, and something comes
back, it is the Web server that provides all the pages you want to view.
If you are in charge of configuring a Web server, there are several things to
keep in mind from the aspect of hardware. First of all, you will have to look at
how busy your Web server will be. For example, the Web server that I have running in my office to provide me and only me with access to my e-mail account
will not be anywhere near as busy as, say, Amazon.com or Microsoft.com.
Therefore, my Web server does not have to have a high-speed connection to the
Internet, really fast network cards, a T1 connection to the Internet, or any of
those things.
If, however, I was configuring a Web server for an e-business site, then all
those things and more would be important. I would need to look at how
many hits my Web site was going to take and plan accordingly. If you are
planning an e-business site, it is always better to be very optimistic than cautious. All you need is to be cautious in your purchase of a Web server, put it
online, and load your Web page, only to find out you have the hottest product on the Internet and your site cannot handle the load. When you are trying
to make money is not a good time to find out your hardware cannot handle
the load.
When you are planning for an e-business site, the thing you should take into
consideration is security. There are many people out there who would love to hack
your site for grins and giggles, or just to get access to your customer list. Losing a
list of customers can be devastating to a company. Losing a list of customers and
their associated credit card numbers can be financial ruin.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

248

Chapter 6

Server Functions

Real World Scenario

So, are you pretty impressed by the different types of servers that are
needed on a busy network? Are you picturing a server room that is slightly
smaller than the footprint of a small office building just to house all these
things? Well, take heartone physical server can handle many of the
servers functions. For example, many companies use a single box as a
DNS, DHCP, WINS server combination, doing some file and print and goodness knows what else.
Not all server types should be combined. Certainly things like application
servers or database servers should be dedicated to a particular task. The
nice thing about deciding when to add one of these server services to your
network is that the decision is pretty clear cut. The need pretty much
stands out, and you can see the service to fill the need.
The other advantage to this smorgasbord of services is that, depending on
your NOS, configuring many of them is a point-and-click proposition taking a
mere matter of minutes. Be sure to spend some time knowing which services
are available and how to make the best use of them, as well as any gotchas that
may come up. The gotchas include things like the configuration of a DHCP
Relay Agent under certain circumstances, or making sure that DHCP is passing
out the right IP address for the new DNS server you just configured.

Conducting Pre-Installation
Planning Activities

Before setting up a new network or adding equipment to an existing network, its important to go through some pre-installation decision-making and
planning. This section talks about the things to consider in the pre-installation
phase of a deployment.

Planning the Installation

You begin by thoroughly understanding and documenting what it is that
youre trying to accomplish. Often, just sitting down and thinking out the
end goal that youre trying to achieve will steer you through myriad vendors

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Conducting Pre-Installation Planning Activities

249

trying to sell you value-add equipment that wont bring you to successful
conclusion of your goal.
Get out a piece of paper and write down what your end-goal is. Then
write down where youre at now and what itll take to get you where you
need to go. In the case of server and network planning, the answer may not
always be so clear-cut. There might be many vendors vying your attention,
for example. Who manufactures the best equipment? Is there a best choice?
Be sure to account for little incidentals in your equipment planning
efforts. For example, one often overlooked item is adequate network cabling
for your new gear. With fiber optic NICs or network gear, your fiber optic
cabling expert will give you a choice of two different types of connectors:
Type SA and Type SC. Youll need to know what youre connecting to, at
both ends, in order to satisfy this request. Power is another consideration.
Some bigger servers utilize 208 voltage instead of the regular 120 volts that
most equipment requires. Its up to you to make sure that the correct lines are
run in order to satisfy the voltage requirements of the new equipment.
Planning involves two things: considering every nuance of a projects installation or upgrade requirements from start to finish, and communication with
the people who can help you figure out the answers to your questions. Vendor
salespeople often bring along a person called a "System Engineer," or SE for
short, to help analyze your particular scenario. The SE is there to answer the
technical questions that you may have that the salesperson isnt capable of
answering. Be sure to thoroughly drill this person with questions you might
have. If youre not satisfied with an answer, tell your vendor so.
Your plan will, when finished, include a project plan with milestones, timelines, dates, and key project elements. Itll also include a spreadsheet that contains the pricing information and itll include information given to you from
your vendor of choice. Management should sign off on the plan or at least
review it for thoroughness. Youll be surprised at some of the intelligent questions a manager might ask about a deploymenteven though the manager
isnt technically inclined. Often a manager looking at a project from a 30,000foot view will see things you might not have thought about while in the midst
of the sea of details.
If youre outsourcing some of the work, youll likely be given a Statement
of Work (SOW). Make sure you review the SOW and are in complete agreement with it before anyone signs it. Contractors are only obligated to live up
to the SOW you sign and may be reluctant to go beyond the letter of whats
in the SOW to help you with problems.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

250

Chapter 6

Server Functions

Verifying the Installation Plan

Once youve got a solid project plan written down, its a good idea to review
the plan to make sure youve covered every jot and tittle. Run the plan by
your vendor of choice to make sure theyre on board with the plan. Also run
the plan by management one more time, just to make sure everyone knows
the nature of work thats going to happen. Complicated SOWs may need to
be reviewed by a company attorney.
With heavily complex or vast plans, it might be good to have another
technical person review the plans for accuracys sake. Better to take a little
time on the front side of the plan and make sure all bases are covered before
proceeding.

The purpose of an installation is to provide you with a blueprint for a complete,

cost-saving, timesaving deployment that impacts users as little as possible.

Verifying Delivery of Correct Components

When the stuff comes in, itll likely be in many boxes. Ive taken delivery on
ATM switchgear that came in several boxes and that I had to put together
into one piece. Servers too. You can pay extra money to have stuff come in
pre-assembled, but you may be satisfied building your own servers or
switches.
Its important that you check the shipping manifesto or invoice against
what youre expecting to make sure you have the correct gear youve
ordered. Remember that computer and network hardware companies ship
out thousands of boxes a day. Its easy to get something confused and,
though it shouldnt happen, it does. Check carefully.

Installing Hardware Using ESD Best Practices

iolating good antistatic practices seems to be the norm for server

admins. Dont ask me why, but most of the admins Ive worked with dont
think a thing about taking a card from a server and traipsing across a carpeted floor with the card in hand to check some Web documentation from
their local machine!

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Installing Hardware Using ESD Best Practices

251

Heres the deal with ESD:



When working on a server, its best to work on it from a grounded

protective rubber mat.

If a mat is unavailable, use an antistatic wrist-strap to ground yourself

to the chassis of the box.

Never carry cards across a carpeted room without first putting them
in a static bag for protection.

Its a good idea to get yourself into the posture of always transporting
cards, DIMMS, or other electronic gear in a static bag whenever
theyre out of the computer.

Use statically protected tools (those whose handles have been treated
with rubber).

Dont leave the cover off of servers when theyre in normal operation.

Dont eat or drink around servers when youre working on them.

Drop a little cola on a server youre building and it may be lights out
for a $20,000 box!

When working on servers, unplug the box before beginning work. Working with servers that have a power source available is stupid, because you
might get electrocuted or you might introduce a static discharge that
could easily toast the innards (the servers innards, that isyours were
toasted when you got electrocuted).

This all seems like pretty sensible stuff. But youd be shocked at how often
these rules are overlooked, usually in the interest of time. Why bother with
a wrist-strap when the last ninety-nine times out of a hundred, youve had
virtually no problems? Well, because theres always that one time when
something might go wrong. Better safe than sorry!

Cutting and Crimping Network Cabling

The biggest issue youll get into if you decide to crimp your own network
cabling is mixing up one of the crossed pairs. There are some key things to
think about if youre going to get into the cable making business:


Adopt a factory sort of standard in your cable making. Always make the
cables exactly the same way each time, being sure to adopt a strict cablemanufacturing regimen. This will help you avoid mis-crossed pairs.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

252

Chapter 6

Server Functions

Do not go beyond Ethernet standards with your cable lengths.

Buy good quality ends and cable. Dont shortchange yourself with
cheap stuff.

My take on cable manufacturing is this: Leave it to the experts, especially

with fiber optic cabling. If you havent been trained to correctly create network cables, youre always better off buying cables that are pre-made, either
in bags from the computer store or from a reputable cabling vendor. Youll
be shocked at how quickly the quality of a network can deteriorate due to
poorly manufactured cables. You might think you can do a better job than
a professional thats in the business, but you might also be introducing
latency and errors into your network simply because youve not correctly
created the cables.

Verifying Power-On via Power-On Sequence

The Power On Self Test (POST) and subsequent power-on sequences that
servers go through can reveal very enlightening information. Its always
wise to watch the computer as it powers up. The first time is useful for verifying that the computers BIOS shows the correct amount of RAM thats
been installed in the computer, and also for validating the correct number
of processors that are installed.
If additional cards such as SCSI or RAID array controller cards are
installed, youll see a BIOS screen show up at power-on. Often the cards
firmware will display a key sequence that you can hit in order to adjust the
cards properties. You might be advised, for example, to hit a Ctrl+M to
enter setup for a particular card. This is true even if the controller is built into
the motherboard.
Some servers come with more than one SCSI controller card and youll see
two instances of the cards firmware show up on screen as the server boots.
Youll have to be aware of the boot order of the cards so that you understand
which one to adjust should you need to adjust its BIOS information.
While powering up, watch for any telltale errors. You might see keyboard
or other errors that appear while powering up. As a general rule of thumb,
its not acceptable for any error to appear at power-up, though Ive been in
situations where I was working with a computer that had a BIOS that
reported an error, but nothing was wrong with the computer. I couldnt
solve the error and the server ran just fine even though the error showed up
at power-up.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Add Processors

253

One of the most perplexing things a new server admin might get into is a
system that boots without video. The reason the problem is perplexing has to
do with the fact that it takes a few minutes for it to dawn on you that the video
is hosed up. The computer is actually going through a wonderful power-on
sequence, but you cant see it because youre plugged into the wrong video
port, or the video card has a problem or some such issue. Listening for the various system beeps will help you understand that the computer is going through
its normal operations but you just cant see them. Its important, of course, to
rectify video issues right away.
Speaking of computer beeps, various motherboards have different ways of
alerting you of normal bootups and of problems. The most common "uh-oh
there is something seriously wrong with the computer" warning is a series of
three or four long tones from the computer. Typically this happens when
youve got mismatched memory in the computer, but Ive heard long tones
with other situations as well. Get to know your servers motherboard!

Add Processors

dding processors to a multiprocessor computer can be easy as pie or

tremendously difficult. Part of the result depends on your understanding of
the computer youre working with and its nuances. In this section we talk
about some of the issues you might face when adding processors.

Perform BIOS Upgrade

Check the Web for the server type youre working with to see if there are
BIOS upgrades that are available and if theyre pertinent to your processor
upgrade. As a general operating principle, its always wise to upgrade a
server to the latest and greatest BIOS before putting the server into production for the first time. However, as time goes on, the servers BIOS gets
out of date, and you might have to investigate upgrading the servers BIOS
before adding processors.
When rebuilding servers that were once heavy production units but have
now been replaced and are being relegated to more minor server duties, its
always wise to upgrade the servers BIOS.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

254

Chapter 6

Server Functions

You can download BIOS upgrades from the manufacturers Web site or
obtain them from the installation CDs that came with the server. Generally
the installation CDs are one or two versions behind current, so its wise to
check the Web.

Remember that when youre adding processors to a computer, oftentimes

each processor has its own DC power conversion card that must be installed
as well.

When moving from a single processor to a multiple-processor computer,

youll probably have to upgrade the computers BIOS to understand and be
able to work with the additional processors.

Performing Physical Housekeeping

cluttered environment is one that invites chaos and leads to poor

network management practices. How so? Because if youre having to work
in a messy server room with lots of cables running everywhere and servers
stacked all over the place, youre not going to be as effective in your problem diagnosis and resolution as you might be if everything were in order
and neatly managed.
Servers should be rack-mounted or, at the very least, able to sit within a nice
rack. Racks provide the ability for you to consolidate your server gear into
"cupboards," if you will, that keep the computers out of harms way, keep
dust from getting inside the computers so easily, and allow you to manage the
cabling going into the rack in a more efficient and neat manner.
Cables should be run through chases under floors and within ducts when
heading toward racks. There are many different ways of running cabling in
neat efficient bundles. Cables should be well marked, on both ends so theyre
easily traceable.
I think nylon wire ties are a curse. I dont like them! Theyre fine as long
as a cable stays operational. But what happens when a cable fails? Then you
have to cut all of the wire ties between the source and the destination, just to
change one cable! What a huge hassle. However, if you dont bundle your
cables using some technique, then youll have a runaway cable mess all over
the floor. Use Velcro ties instead and youll find that you can bundle your
cables without the permanence of a nylon tiethus facilitating ease of management when (not if!) a cable fails.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Establishing Remote Notification

255

Performing Hardware Verification

Hardware verification amounts to two separate concepts: that of maintaining a hardware asset inventory, and also of keeping track of the various
firmware and BIOS version numbers for a system.
With most network operating systems, theres usually a recommended supported list of hardware that can be utilized within a given complete system. Its
vitally important that you pay close attention to this "blessed hardware" list
predominantly because if you dont, theres only so far that a manufacturers
support person will go with you should you experience problems. In a Windows-based system, Microsoft calls this list the Hardware Compatibility List
(HCL). Other manufacturers have comparable lists. Do not try to install an
NOS on a computer whose combined parts dont fit within a supported configurationnot unless youre fond of troubleshooting on your own, that is.
Take down complete system information. Include the brand name and
BIOS/firmware version numbers of the cards in the system. Keep a good
record of each computer in the server farm. Consider routinely updating a
computers BIOS.

Establishing Remote Notification

ith many different server manufacturers you have the option of purchasing a separate add-on card that allows you to connect a phone line to the
computer and then be able to dial in to the computer when you want to check
its status. The computer can also phone out with alerts that revolve around
something physically going wrong with the computer. For example, suppose
that there is a piece of paper plastered against the fan and the computer is heating up. It would be nice if you could be paged or somehow notified that this
event occurred. With these separate add-on cards, you can arrange this kind of
notification, independent of the NOS.
There are alerting software packages that are available for a server NOS
that can allow you to configure advanced notification options. For example,
suppose that youre interested in monitoring a given daemon, service, or
NLM to assure that its consistently up and operational. You can apply software solutions to such a need and obtain alerting in this way. You can set up
a system to send a message to your pager, to email you or, in some cases, to
call you.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

256

Chapter 6

Server Functions

Note that the above two paragraphs describe two very different methods
of handling remote notification. The first describes the very base functions of
the system, independent of the NOS installed on it. The second describes
software that isnt necessarily interested in whether the NOS is running, but
is concerned with given applications and events.
You can combine the two monitoring and alerting processes using
software called Enterprise Management software. Companies such as
Computer Associates with their Unicenter TNG, or IBM with its Tivoli
offering, provide very robust and granular management tools that can
continually observe and alert on virtually every facet of a servers operationfrom hardware failure to service management.

Summary

hew! That was one extensive list, wasnt it? We have looked at
about every way imaginable that a server can be used. I understand that we
did not do an in-depth look at any of the server types, but entire books can
be written on topics like DNS, DHCP, and Web servers.
This is one of those chapters that are target-rich for test writers. There is
a lot here to choose from in the way of questions. To make it worse, so many
of these servers do close to the same thing or look and sound so much alike.
For example, if I asked you what type of server resolves NetBIOS names to
IP addresses, it would be WINS. Change one word and the answer would be
DNS. Hey, those are two great questions for the assessment test or for the
review questions.
We also talked in this chapter about pre-installation planning considerations. Especially with larger server rollouts, you should prepare a project plan
that details how youre going to accomplish your goal and communicate it
with others.
We mentioned that you should follow good ESD practices when working
with servers.
We also talked about remote notification cards and software for servers
and mentioned that good housekeeping when racking and cabling servers is
highly important. In the cabling realm, we noted that cables that you manufacture yourself have a greater propensity for failure than those you obtain
commercially. Especially important if youre "home-growing" your cables is
to make sure that the crossed pairs are the correct pairs. But overall, its to

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Summary

257

your benefit to obtain your cables from a reputable cabling vendor or

through a catalog or supply store.
We also talked about updating the BIOS on computers, especially when
youre getting ready to add processors. And we mentioned that its important
to watch a servers power-up sequence, paying special attention to errors that
might pop up and watching the add-on cards BIOS information as it displays.
Next up, we will start by looking at how some servers are used and then
move on to backup, restore, and disaster recovery, and look at ways to
physically secure a server.

Exam Essentials
Know what SNMP does SNMP tracks events on the network. This can
include things like servers coming online or going offline. It can also track
other events like logons or logoffs.
Know why DMI is used DMI is used to remotely manage a users desktop.
Know when to use a gateway A gateway is used to allow communication between dissimilar systems. For example, an SMTP gateway will
translate information from the e-mail system you are using to the protocol
used by the Internet.
Know when to use a router A router is used to efficiently manage communication between computer networks.
Know when to use a bridge A bridge is used to manage communication
between systems on a network segment.
Know when to use a firewall A firewall is used to protect a network (or
even a network segment) from intruders.
Know when to use a proxy server A proxy server is used to enhance
performance of accessing Web pages from the Internet or from an
intranet. It can also be used as a rudimentary type of firewall.
Know when to use a database server A database server is used to manage
large amounts of data stored in a database format.
Know when to use client/server Client/server communication allows a
particular application to fulfill a clients request, rather than having the
client going out and searching through databases. This usually enhances
the security of the database from unwanted corruption.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

258

Chapter 6

Server Functions

Know when to use an application server Application servers may have

several different definitions, but in this case, think of a Web server providing
application support to your organization.
Know when to use a mail server Mail servers are used to provide your
organization with e-mail capability. This can be Internet mail, or a type of
groupware like Lotus Notes, Microsoft Exchange, or Novells GroupWise.
Know when to use an FTP server FTP servers are used to provide files
to users, or to provide a collection point for gathering files from users.
Know when to use an SNA server SNA servers are used to provide
access to IBM mainframes.
Know when to use a RAS server RAS servers provide dial-in, dial-out
support.
Know when to use a file and print server File and print servers give
users access to data files and the ability to print them on shared print
devices.
Know when to use a fax server Fax servers are used so that users will
not have to leave their desks to send or receive faxes.
Know when to use a DNS server DNS servers are used to resolve userfriendly DNS names to computer-friendly IP addresses.
Know when to use a WINS server WINS servers are used to resolve
user-friendly NetBIOS computer names to IP addresses.
Know when to use a DHCP server DHCP servers are used to distribute
TCP/IP addresses as well as other configuration information.
Know when to use a Web server Web servers are used to provide access
to a variety of different document types, like on the Internet or on an
intranet.
Be familiar with the pre-planning stages of a deployment Understand
the various documents that youll want to have in place and the major
players that will be involved in going forward with a rollout.
Use good ESD practices when installing new server gear Dont get
cocky or let your guard down when working with cards in servers or
workstations.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Summary

259

Key Terms
Before you take the exam, be certain you are familiar with the following terms:
application server
bridge
client/server computing
cold start
community
Component Interface (CI)
data warehousing
database server
Desktop Management Task Force (DTMF)
Domain Name Service (DNS) Server
Dynamic DNS (DDNS)
Dynamic Host Configuration Protocol (DHCP) Server
Exterior Gateway Protocol (EGP)
fax server
file and print server
File Transfer Protocol (FTP)
firewalls
full zone transfer
gateway
HyperText Transfer Protocol (HTTP)
incremental zone transfer
item
local area network (LAN)
location
mail server
Management Information Base (MIB)

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

260

Chapter 6

Server Functions

Management Information Format (MIF) Database

Management Interface (MI),
media access control (MAC)
network interface card (NIC)
non-routable protocol
Open System Interconnect (OSI)
portal
primary zone
process
Proxy Server
Remote Access Services (RAS) Server
Reverse Proxy Server
router
scope
secondary zone
Service Layer
Simple Mail Transfer Protocol (SMTP)
Simple Mail Transfer Protocol (SMTP)/Multi-Purpose Internet Mail
Extensions (MIME)
Simple Network Management Protocol (SNMP)
SNMP Agent
SNMP Manager
Statement of Work (SOW)
System Network Architecture (SNA)
warm start
well known port
Windows Internet Name Service (WINS)
zone
zone transfer

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

261

Review Questions
1. Which server type would you generally associate with things like

POP3 or SMTP?
A. DNS
B. DHCP
C. E-mail
D. Application
E. Web
2. Which protocol would you associate with server and network

management?
A. DMI
B. DNS
C. ESPN
D. SMTP
E. SNMP
3. Which of the following are components in an SNMP implementation?
A. Manager
B. Hosts
C. DMI
D. Agents
E. IP
4. From the list below, select some pre-planning activities.
A. Preparing a project plan
B. Speaking with various entities involved with the project
C. Updating the computers BIOS
D. Checking hardware compatibility lists

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

262

Chapter 6

Server Functions

5. What are some practical ESD practices to put into place?

A. Use static bags.
B. Only work in ESD-ready rooms.
C. Wear wrist-straps.
D. Unplug servers.
6. Ians trying to install a second processor in his server, but the BIOS

only recognizes one at power-up time. What could be the problem?

A. Computers not rated for multiple processors
B. DC power supply for second processor missing
C. BIOS needs to be upgraded
D. BIOS needs to be configured for second processor
7. Elizabeth purchased a server that includes an option management card

allowing her to directly dial the server, independent of the motherboard, or she can have the server send out alerts using the phone line
in the card. Now Elizabeth is configuring the card so that it pages her
when theres an issue at the hardware level. What feature is Elizabeth
implementing with this card?
A. Remote control
B. Remote notification
C. Remote diagnostics
D. Remote authentication
8. Francois has been hired as the new server administrator for a small

company that manufactures and sells bicycle equipment. He has

started a process of going through his servers and verifying the hardware that they utilizegathering serial numbers, sizes, types, that sort
of thing. What sorts of things can Francois use this information for?
A. DNS
B. Upgrade information
C. To make cost justification documents
D. To provide inventory information to stakeholders

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

263

9. Why should network cabling be run through innerduct and banded

with Velcro or nylon to rails inside racks?

A. Keeps people from stepping on the cable and destroying it
B. Keeps cables from coming unplugged in devices
C. Ensures that cables wont get into crosstalk situations
D. Presents a nice professional appearance
10. For security purposes, SNMP Agents can be gathered into:
A. Subnets
B. Communities
C. MIBS
D. MIFS
E. SNMP
11. Erin is trying to add a processor to her server to make it a dual-processor

computer. She has validated that the speed, stepping, and L2 cache size
are correct. She has the correct parts, but when she installs the processor
the computer wont recognize it. Whats the likely problem?
A. She has to enable the second processor port in the system BIOS.
B. She has to upgrade the system BIOS.
C. She has to upgrade the first CPUs BIOS.
D. She must first reboot the server.
12. Whats one of the best ways to tell if the addition of system memory

or an additional processor was successful?

A. Check the system BIOS.
B. Check the NOS device statistics.
C. Watch the power-on sequence.
D. Review the error logs.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

264

Chapter 6

Server Functions

13. John is preparing a server that will be used as the companys new

e-mail server. Because he gets many phone calls in a day, he wants to

install and configure the server at his desk. What precautions should
he take?
A. Keep new server gear on a rubber mat.
B. Wear a wrist strap.
C. No precautions are necessary.
D. The server should be plugged in so its connected to an earth

ground at all times.

14. What important task should you perform when receiving new

equipment?
A. Power up and test new equipment before putting it into production.
B. Validate the order against what was shipped.
C. Jot down all serial numbers.
D. Read the manuals.
15. Youve written out a plan for a brand new server installation. This

server is more technically challenging than any youve set up before.

What should be your next course of action?
A. Have someone else validate your plan.
B. Check Web documentation for clarification on any issues.
C. Deploy the new server.
D. Read the manuals for the server.
16. Jason has convinced his boss that the company can save money by

letting him purchase a reel of Ethernet CAT 5 wiring, some RJ 45

plugs, and a gun so he can do his own cable creating. What must Jason
watch out for as he ventures out into this new methodology?
A. Violation of Ethernet cable length rules
B. Getting his crossed wire pairs wrong
C. Loose connections
D. Wrong crimper gun

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

265

17. Name some server devices that have a BIOS that you must consider

when thinking about upgrades.

A. SCSI adapters
B. RAID array controller cards
C. Motherboard BIOS
D. Serial/parallel cards
E. Video cards
18. Name some components of a good installation plan.
A. Ensuring adequate power
B. Preparing for network connectivity
C. Formulating an IRQ chart
D. Obtaining a good warranty
19. Why cant you run an internal cable through a slot in a computer to an

external device?
A. You cantheres nothing wrong with it.
B. Violates good ESD practices.
C. Cables arent made for that.
D. External devices cant take an internal connection.
20. Stan has done some work on his server and now it appears to have no

power at all. How can Stan verify that the servers getting power?
A. Check the cooling fan.
B. Test it with a circuit tester.
C. Plug the computer into a different power socket.
D. Check the panel display.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

266

Chapter 6

Server Functions

Answers to Review Questions

1. C. SMTP and POP3 are protocols associated with electronic messaging.
2. E. SNMP is Simple Network Management Protocol. The key words

here are server and network management.

3. A, D. SNMP depends on Managers and Agents.
4. A, B, D. You should definitely update the computers BIOS, but that

happens much later on after youve received the gear and youre prepping it. Its very wise to ascertain ahead of time that youre purchasing
the correct complement of components for the server. You should also
develop a project plan that details how youre going to obtain, build,
and deploy the server and you should be sure you communicate with
other stakeholders involved with the new system.
5. A, C, D. While its great to work in an ESD-ready room, you wont

always have that luxury. Use static bags when transporting cards.
Wear ESD wrist-straps when working on a servers internals and
always unplug the server before working on it.
6. B, C. Its not likely that youd be able to buy a second processor for a

computer thats not rated for multiprocessor capacity, so A is out. It is

possible that you forgot to put in the additional DC power supply. Its
also quite possible that the computers current BIOS doesnt support
multiprocessing and that youll need to upgrade the BIOS before going
on. You usually dont have to adjust any of the BIOSs configuration
properties when installing a second processor.
7. B. Elizabeth is utilizing the remote notification feature of the card.

Note that some NOS software can be configured to send alerts out to
administrators when certain key events occur. Setting up alerts on
servers is something administrators should be responsible and diligent
about doing.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Answers to Review Questions

267

8. A, B, D. You could certainly use hardware verification information to

tell you the kind of device youre replacing in a situation where a device
has failed. You could also use the information to help you make decisions about an upgrade and you could certainly key the information
into a spreadsheet to provide inventory information to others. However, as new hardware products come out, they often overshadow the
old in terms of lower pricing and higher output or efficiency.
9. A, B, D. You have no control over the way cables are run inside walls

so you dont know if you may have a cable running parallel to a phone
line (thus generating crosstalk) or not. But by keeping the cable runs
neat, you can ensure that people dont step on the cables and that the
cables dont become unplugged from devices. It also creates a nice,
neat, professional appearance.
10. B. SNMP Agents can be gathered into communities for security

purposes.
11. B. You might run into a circumstance where an older BIOS doesnt

recognize that its working in a multiple processor system. Upgrading

the BIOS allows the system to see the second processor port and the
new CPU will install just fine.
12. C. If youve installed a new or second processor or added system mem-

ory, you should instantly see the changes on screen at startup.

13. A, B. Server admins are notorious for working on gear thats not ESD

protected. Shame on them! In a scenario such as the above, you would

definitely want to consider keeping the new gear on a rubber mat and
wearing a wrist strap. Some rubber mats have a snap on them that
allows you to snap your wrist strap directly to the mat rather than
somewhere on the computer.
14. B. Its important to check what you ordered against what was sent.

Hey, computer vendors make mistakes just like everybody else! Why
find the problem at 2:00 A.M. when youre ready to install the server?
15. A. You should have someone else who understands the nuances of

what youre doing validate your plan. If theres no one available

locally to look at your plan, consider asking the vendor to take a quick
look at it. Better safe than setting up a half-baked server installation.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

268

Chapter 6

Server Functions

16. A, B, C. Improperly crimped cables can create havoc on a network.

The most common problem hell run into will be cables that dont
work because he got the crossed wire pairs wrong. This is work thats
better being outsourced or purchased over the counter.
17. A, B, C, E. SCSI adapters, the BIOS on the motherboard (the system

BIOS), RAID cards, and video cards all have BIOS firmware that
might need to be upgraded. Serial/parallel cards generally do not.
18. A, B, C. A good installation plan will help you think about things you

might not have considered had you not taken the time to sit down and
think out what youre going to do. For example, making sure you have
adequate power in the server room is an installation must. Also making
sure you have network connectivity is important and seeing to it that
you use the correct IRQs for all devices might also be a good thing to
check during installation. The warranty doesnt have anything to do
with the installation.
19. B. Items C and D are correct, but Ive seen admins do some creative

splicing and get an external connection hooked to an internal cable. The

bigger problem is that the internal cable isnt shielded the same way an
external cable is and youre violating good ESD practices. Besides that,
it looks really amateur.
20. A, D. You might run into this problem. You make a change to a system,

particularly its motherboard, and the system appears to completely stop

functioning. Take a look at the cooling fan. If its running, you know
youve got power to the power supply and the change you made didnt
blow it out. If the front panel has an LED display, you can tell that way
as well.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Chapter

Server Etiquette
SERVER+ EXAM OBJECTIVES COVERED IN
THIS CHAPTER:

1.2 Install hardware using ESD best practices (boards, drives,
processors, memory, internal cable, etc.).


Mount the rack installation.

Cut and crimp network cabling.

Install UPS.

Verify SCSI ID configuration and termination.

Install external devices (e.g., keyboards, monitors,

subsystems, modem rack, etc.).

Verify power-on via power-on sequence.

3.1 Perform full backup.



Verify backup.

4.1 Perform regular backup.

6.2 Use diagnostic hardware and software tools and utilities.


Identify common diagnostic tools across the following OS:

Microsoft Windows NT/2000, Novell Netware, UNIX, Linux,
IBM OS/2.

Perform shut down across the following OS: Microsoft

Windows NT/2000, Novell Netware, UNIX, Linux, IBM OS/2.

Select the appropriate tool.

Use the selected tool effectively.

Replace defective hardware components as appropriate.

Identify defective FRUs and replace with correct part.

Interpret error logs, operating system errors, health logs,

and critical events.

Use documentation from previous technician successfully.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Locate and effectively use hot tips (e.g., fixes, OS updates,

E-support, web pages, CDs).

Gather resources to get problems solved: identify situations

requiring call for assistance; acquire appropriate
documentation.

Describe how to perform remote troubleshooting for a wakeon-LAN.

Describe how to perform remote troubleshooting for a

remote alert.

7.1 Plan for disaster recovery.



Plan for redundancy (e.g., hard drives, power supplies, fans,

NICs, processors, UPS).

Use the technique of hot swap, warm swap, and hot spare to
ensure availability.

Use the concepts of fault tolerance/fault recovery to create a

disaster recovery plan.

Develop disaster recovery plan.

Identify types of backup hardware.

Identify types of backup and restoration schemes.

Confirm and use off site storage for backup.

Document and test disaster recovery plan regularly, and

update as needed.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

n this chapter, we are going to start by taking a closer look at several

different types of application servers. We touched briefly on one of the types in
the last chapter, a dedicated application server, but in this chapter we will also
look at a distributed application server and a peer to peer application server.
After we get done rounding out our discussion of server types, we will
take a look at the care and feeding of servers. This will include several of the
most important tasks a network administrator can do to protect the server.
Finally we are going to look at ways to properly shut down a server and how
some of the commands work to protect the operating system and the data
files that may be on the server. So, that is what there is on tap for you in this
section. Not as much sheer overwhelming information as in the last chapter,
but good stuff nonetheless.

For complete coverage of objective1.2, please also see Chapters 1, 6, and 8. For
complete coverage of objectives 3.1 and 4.1, please also see Chapter 12. For complete coverage of objective 6.2, please also see Chapter 13. For complete coverage of objective 7.1, please also see Chapters 1 and 12.

Application Servers

n order to figure out what the different types of application servers

do, you first must understand what an application server really is. The
application server is a server that runs in a client/server-based environment. This server will be running one or more than one application that
can be shared by the clients. In addition, the server will be doing some of
the data processing tasks. The application server model provides for

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

272

Chapter 7

Server Etiquette

faster data processing and faster throughput of information provided to

the client stations.
The application server can be part of a three-part application:


Graphical User Interface (GUI) server

Application or business logic server

Database server

The first part of the application is usually a Web browser user interface
that can be accessed from a remote workstation. The middle section is an
application or applications housed on a local area network or on an intranet
server. Finally, the last part is the database and transaction server. This can
be housed on a larger server, up to and including a mainframe.
As was mentioned in the last chapter, Server Functions, the entire purpose of an application server is to deliver applications to the network users.
Take a look at Figure 7.1 to see how this all lays out.
FIGURE 7.1

Application server layout

Internet
Web-based user

Web Server

Application Server

Database Server

You can see how the end user accesses the information using a Web
browser or a thin client. The first place the signal hits is the Web server, which
passes the information along to the application server. The application server
uses the database server to either retrieve information that the client needs or
to store information from the clients.

A thin client is a specially designed client that allows you to access information
from a system like Microsofts Terminal Server or Citrix Metaframe. It sends
and receives keystrokes, mouse movements, and video updates. All processing
is done at the server.

There are several ways the application servers can be configured. Lets
look at some of them.

Dedicated Application Server

The name dedicated application server says it all. This is a server that is
dedicated to providing the application services for the client/server-based

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Application Servers

273

environment. This can be a server that handles millions of transactions

every day for a huge e-commerce site. It could also be something as routine
and mundane as an e-mail server. In either case, the server is dedicated to
running the application and storing the data.
With a dedicated application server, all requests come directly to the application server. The application server then can process the request and access the
database server. By having the application server be dedicated, you are reducing
administrative costs. Since there is only one server (or cluster of servers), there
only has to be one administrative staff trained and dedicated to managing the
server. Compare this to the distributed application server mentioned next, where
you will need a support staff at several locations.
Hardware costs, while not cheap by any means, will also be less. You will
be able to save money by purchasing a single unit rather than installing and
configuring multiple smaller units.

Distributed Application Server

The easiest way to describe what happens with a distributed application
server is to show you Figure 7.2.
FIGURE 7.2

Distributed application server

Database Server

Local site
LAN
Application Server
Application Client
Server

Client

Client

Application
Server

Remote site
LAN

Application Client
Server

Copyright 2001 SYBEX, Inc., Alameda, CA

Client

www.sybex.com

Client

274

Chapter 7

Server Etiquette

In this case, there are application servers running at various locations

throughout the enterprise. All of those servers are accessing information from
a centrally located database server. In this case, the client/server applications
can be run closer to the client, eliminating wide area network traffic. Instead
of each client having to send information over the wide area network link, only
the application server has to traverse the network, and then only for information it does not have access to locally.

Peer to Peer Application Server

Are you a gamer? If you are, you may have used a type of peer to peer application server. Suppose that you and your favorite Internet gaming buddy are
playing a game. There may be information or applications running on each
of your systems, which turns this into a peer to peer application server. In
each case, the servers may be asking each other for information as the screens
get redrawn.

Real World Scenario

The decision to use a dedicated application server, a distributed application server, or a peer to peer application server is really not a conscious
decision. The application that is going to be running on the server will
determine the type of server, and in some cases the role will evolve or
change over time. For example, today I may have a dedicated application
server running an e-mail program. As my e-mail system grows, I create
another mail storage facility on another server in the same room, but run
the mail program on the original server. I have now changed my dedicated
mail server into a distributed application server.
Dedicating servers to a particular task is usually determined by the
amount of work the server is going to do. As I have mentioned before,
most servers provide multiple types of servicesfor instance, your mail
server can also be providing DHCP services.

Disaster Recovery

While the test objective lists disaster recovery after backup, I think it is
important to reverse the discussion here. You see, some of the decisions that you

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Value of Services

275

make while you are deciding on disaster recovery will have a definite impact on
your backup strategies, so lets get this discussion out of the way first!
Disaster recovery has evolved into almost an art form. The premise is really
simple. If something were to happen to a computer or group of computers,
how long would it take before the services offered by those computers would
be available? What steps have to be taken to bring the restoration of services
into play? That is all there is to itsort of.
Lets take a more in-depth look at what some of those criteria mean to the
average business. First of all, the question starts out, If something were to
happen to a computer or a group of computers.... Your immediate reaction
should be, What could possibly happen to them?
If you start by identifying disasters, then you start to determine what you
can do to prevent the disaster or recover from it. For example, lets say you live
in Jacksonville, Florida. You have a generic computer. Now, what disaster can
befall that device? First of all, it might be something common like a failed disk
drive. Or, the disaster may be something more serious, like a communication
link between two key hubs is down or the computer failed because the building
that housed it was hit by a hurricane and the building is now underwater.
In a good disaster recovery plan, all of these things have to be taken into
consideration.
So, if I were to come in and start the disaster planning process, I would
start by making a list of the common disasters that can befall a computer
room: The power goes out, lightning strikes, a disk drive fails, the building
burns down, the entire block burns down, the entire town floods, or there is
a war. Once I have a list of potential disasters, I can move on to the next step.

Value of Services

ow we are getting into areas where some assumptions will have to

be made, where there probably isnt hard and fast information available.
This is called establishing the level of risk, and it turns the process from an
interesting mental exercise into something much more concrete. Now you
have to start assigning dollar values to the services and information your network protects. How tough is this? Well, for example, what is the value you
can place on data? That depends on the kind of data. If you ask me to place
value on the data contained in a spreadsheet that assigns parking places to
the 25 employees of the company, that is not going to be valued at much. But
if you ask me to put a valuation on the information stored on a server that
is used as an ISPs main Web server, and it hosts dozens of e-business sites for
companies around the world, suddenly the value of that system has increased
dramatically. How can you decide?

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

276

Chapter 7

Server Etiquette

Chances are, you cant decide. You probably dont have enough information
at your fingertips to make those kinds of decisions. This is going to have to be
the decision of a committee and that committee should have representatives
from all aspects of the business. When you put your disaster recovery team
together, be sure to include all facets of the business, from the management folks
on down. Make sure also that everyone has the freedom to say what is on his
mind. Too often, people think they know how something is used, and may have
no clue about its real value to the company. I am sure there is no one like that
where you work, but there have been several people in my past that were in the
position of authority who were completely in the dark about the way things
really worked.
Now that you have the committee together, it is time to meet and look at
what would happen if a resource suddenly disappeared. What would be the
impact on the company? Here is an example. Earlier in the book, we talked
about different types of servers. Lets assume that the company you are
working for printed out mailing labels for customers. If your main database
server went down, with thousands and thousands of names on it, it may put
your company in difficult straits, depending on how long it took you to
restore the information. If it was completely and totally out of commission
and you suddenly found there were no copies of the data and no backups,
your company would effectively be out of business.
Contrast that to what would happen if the file and print server went down
in the word processing area. Depending on the information typed on a daily
basis, your level of pain may range from mildly inconvenienced to seriously in
a world of hurt. Start the committee meeting by objectively looking at your
resources. Once you have listed the resources, then it is time to figure out how
long the company can live without the resource and what the company would
be willing to do to protect the resource. Finally, you and the committee have
to determine if this tack is financially and technically feasible.
If your company is like most of the ones I have dealt with, there is no such
thing as disposable, noncritical data, even if the data consists of agendas
from meetings that were held five years ago. It obviously is important to
someone; otherwise it never would have been saved to the network. Since it
is important, it has to be protected. Now, the real world steps in. How much
would the company lose per hour (or in some cases, per minute) if the service
or information were not available? Once that has been determined, how
much is the company willing to spend to protect it?

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Value of Services

277

Lets look at a stock brokerage firm. How much would the company lose
per minute if its electronic link to the stock exchanges went down? First,
there would be the tangible losses. For example, I wanted to buy 100 shares
of ACE Company at $25 a share. When the link went down, the stock had
just hit the trigger point, but by the time the link was back up, the stock was
trading at $25.50 and the buy order went unfilled. That may have cost the
company $75. Then there is the loss of trust from the client. I call, wondering
why my order wasnt filled, and you tell me it was due to a computer glitch.
I am going to be looking for a brokerage company with a better computer
department. Multiply that story by several thousand customers and you can
see we are talking some big numbers. Some brokerage companies will handle
millions of dollars an hour. If the computer services are not available, that
can add up to real money really quickly.
Once you have made your list of critical services and decided how
much those services are worth, it is time to figure out what it will take to
provide those services in a manner that meets the corporate needs. It is
not uncommon for companies to reevaluate the value of services based on
the cost of providing the mission critical disaster recovery services. Lets
take a look at an example. In any company, it is not uncommon for
fiefdoms to spring up. During this process, you will see people become
very protective of their data. You may also see some amazing transformations. Take the person who is in charge of the database that manages
the parking permits. She may come up with a very convincing argument
as to why that information cannot be down for even 1/10th of a second.
The argument may be so good that you may want to suggest that person
move into the companys marketing department. Anyway, once the information has been given a high priority, it is up to you to protect it 24 hours
a day, 7 days a week. This is where you suggest that there be a mirrored
server placed off-site to accomplish the task. Cost of the server, $7,500.
Cost of the facility to hold the server, $10,000 a year. Cost of a leased
line, $1,000 a month. Cost of technology to mirror servers, $10,000.
Suddenly someone takes a close look at the costs and determines that the
company can put up with a lot of parking lot chaos for that amount of
money. Can you see how this is an ever evolving document?
Once the committee has completed the list, it is time to take the list to the
next level. The next level is where you determine how you are going to identify
that a risk has occurred, and what are you going to do to recover. Once again,
some of this is exceptionally easy. Having a dozen firefighters outside the corporate headquarters sending streams of water into the building is a pretty good
indication that there is a fire. But how do you tell that someone from outside

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

278

Chapter 7

Server Etiquette

the company is currently accessing your customer database? That is not nearly
so simple. What could be even more difficult to detect would be if someone
from inside the company who was not supposed to have access were currently
accessing your customer database. Each of these occurrences can be a disaster
just waiting to happen. Protecting against it can appear easy, but determining
the fact that you have had a disaster can be difficult.
To make this even more of a challenge, you are going to have to enlist the
help of people who are not normally known as IT professionals, including
management. Think about it. The data that is probably the most vulnerable
to attack is the data that the upper echelons of managers will be seeing. In a
normal organization, whatever your definition of normal is, the higher up
the corporate ladder an individual goes, the less she knows about the nuts
and bolts of the network. So there will be some training involved. The training will include things like being aware of things around you. If something
in a document does not look right, how can you check to see if it has been
compromised?

Putting the Recovery in Disaster Recovery

Once you have determined how much you can do to protect the information, it is time to examine what you can do to recover from a disaster
when equipment fails or is destroyed and/or data is lost. In previous chapters
we have covered some of the steps you can take to ensure against data loss
through equipment failure, but what if your equipment is destroyed?
What would you do if your pager went off in the middle of the night and
you were told to get to the office as quickly as possible? Youd go. Imagine
the look on your face when you rounded the last corner and pulled into the
parking lot only to find dozens of fire trucks fighting a massive blaze. Could
your company survive this type of disaster? How long would it take them to
be back online and ready to work? For some companies, the time period
would be months. For other companies, the time period might be days. Some
companies may have switched their computer operations to a different site
and be online before the first fire truck arrived on the scene. It all depends on
how deep your pockets are, and what the cost of failure would be.
Lets take a look at this from the server room point of view. This probably
isnt the first time I have used dirty words in this book, but it may be, so prepare
yourself. You are going to make your life much easier if you have documented
your current network. Now, you and I both know that the term documentation
is usually at the top of your list of things to do. And it stays there and never gets
crossed off, mainly because you are too busy with life to do what you have to do.
Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Putting the Recovery in Disaster Recovery

279

If you are serious about disaster recovery, there are four terms that should
be your mantra:


Create a plan for every eventuality.

Document everything.

Back up everything.

Store it off-site.

Proper Planning Prevents Poor Performance

Think of all the time you spend researching purchases, deciding on the different components that make up your servers and workstations, getting the
proper configuration, and then sticking with it. Wouldnt it be wonderful if
you could avoid doing that again? That is what documentation will help you
with. Lets assume your building is gone. You have no spare inventory for
computers and servers and your bosses have just located space where you
can start installing a network as soon as you get the boxes. Wouldnt it be
wonderful to be able to call your vendors and give them the exact specifications of the servers you had, so they can be replaced? The same is true with
workstations.
Besides the hardware and software specifications, there are other things
that should be included in that disaster recovery documentation. These items
are the mundane things of life, like phone numbers. Think how much time it
would save if you could go to the off-site storage location and grab a notebook
that had all your key vendors emergency numbers in it. This document would
include things like who to call to order the T1 line or wireless connection for
the network, who to call to replace the routers, and who to call to replace the
servers. Each entry should have any other key information like account numbers included, anything that will speed up the process of talking on the phone
with the vendor. It is true it only takes a few minutes to verify your customer
name and user ID, but when you are making 50 phone calls, those few minutes
can add up.
Now that we can assume the equipment is on the way, lets hope there is
data to put on it.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

280

Chapter 7

Server Etiquette

Back Up, Back Up, Back Up

It is amazing how many people just dont understand how important this
task really is. Of all the things that you do, this is one of the most important.
When I teach, I like to tell my students some of the rules of network computing.
You know the ones:


End users lie.

The higher you ascend the corporate ladder, the less you know about
computingbut you get better grades of equipment.

You can tell the skill level of a network administrator by his level of
paranoia.

The last one is really my favorite. It points out one of the all time truisms
of server administration: Murphy was an optimist. So, in order to make sure
that Murphys Law doesnt come back and bite you when you least expect
it, you have to make sure that your data is protected.
Earlier in the book, we talked about things like mirrored and duplexed
drives, as well as RAID technology, but there isnt much that can beat a good
old-fashioned full backup, checked and tested and stored off-site to ease the
pangs of stress in a network administrators heart.
There are several different kinds of backups, from the old COPY/XCOPY
command, to full, incremental, and differential. In this section, we will take
a look at the types of backups and some of the occasions to use each. We will
also look at some of the inherent dangers in the process.

Types of Backups
When you start to lay out your backup strategy, there are several things to
keep in mind. Look closely at the server you are going to design a backup strategy for. How often does the information on that server change? That can
impact the strategy. How mission critical is the server, really? That will have
an impact on the backup strategy. Perhaps the most important consideration
is, How fast is the backup device you are using to make your backups and
how much data do you have to back up?

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Back Up, Back Up, Back Up

281

Backups and Server Types

Why would the type of server impact the type of backup strategy you would
use? Lets look at a few examples. First of all, lets take a look at the example
of an e-mail server. Your typical e-mail server is nothing more than a combination database server with some client/server application server thrown
in for good measure. All that other stuff about moving e-mail is just a disguise for making lots of entries, usually small, in databases. These databases
are constantly being updated, and you are tasked with taking a snapshot of
them at a particular time of day.
The size of the e-mail server can vary considerably. If your e-mail implementation is using POP3, you probably have a relatively small data store on
the server, with all of the messages being pushed down to the clients. If you
are using something like Microsoft Exchange, Lotus Notes, or Novell
GroupWise, your message store could be very large; when it comes time to
choose a backup environment then, you will want hardware and software
that will back up large, open databases very quickly.
Compare this to a server that is used as an application deployment server.
The data on this server may be accessed daily, but not changed in weeks. There
are large amounts of data to be backed up, but the criticality of the data being
backed up nightly is not the same as an e-mail server. Finally, you have the file
and database servers. Here, a large amount of information is being changed on
a daily basis. This is the lifeblood of the company. If something happens to one
of these servers and data is lost, there could be very serious consequences.
The types of backups and the frequency of these backups will have to be
determined based on the individual characteristics of the servers.

Types of Backups
Generally, there are four types of backups:


Copy Backup

Full Backup

Differential Backup

Incremental Backup

Each of these types is used to provide protection in different circumstances. I

will start by looking at the type that is the most likely to be consistently
overlooked.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

282

Chapter 7

Server Etiquette

Copy Backups
This is usually the quickest type of backup, the kind of down and dirty
backup that protects a small amount of data for a short period of time.
Suppose that I am going to be making some changes to someones computer
that will affect her POP3 e-mail client, like Outlook Express. I know that no
matter how careful I am, something I do may trigger an unexpected process
that will delete all the address book entries or messages. I know this, because
it has happened to me before. Usually when this happens, the end user loses
most, if not all, of her sense of humor. An easy way to protect against this is
to make copies of the affected files before starting to do any work with them.
It is quick, it is down and dirty, but it at least gives you the opportunity of rolling the procedure back to the time when everything worked. When in doubt,
back it up.
Full Backup
A full backup does just what the name implies. It backs up all the information on the disk to an offline device. This device can be a tape, an optical
drive, or a CD if you have a small amount of information. All information
from one medium is copied to another medium, regardless of when the data
was last accessed. When a full backup or normal backup is done, it will reset
the archive attribute on each file. The attribute will be changed back to
archive if the file is changed in any manner. The full backup and verification
will take the longest to perform.
Incremental Backup
An incremental backup will back up only those files that have been changed
since the last backup. This means that if I do a full backup Sunday night, and
an incremental backup Monday, the incremental will back up all the files that
have been changed on Monday and reset the archive attribute, indicating the
file has been backed up. If you run the incremental backup again on Tuesday,
all those files that have been modified on Tuesday will be backed up, the
archive attribute will be reset, and so on. The incremental backup will take the
least amount of time to run each and every night.
Differential Backup
A differential backup is a backup of only those files that have been modified
since the last full backup was performed. To put this in perspective, lets say
you do a full backup every Sunday night. Using a differential backup on
Monday night will back up all files modified on Monday. Using a differential
backup on Tuesday will back up all files modified on Monday or Tuesday.
Using a differential backup on Wednesday will back up all files modified on

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Back Up, Back Up, Back Up

283

Monday, Tuesday, or Wednesday. As you can see, the differential backup

will take progressively longer to run each night.

Backup Strategy
Once you have determined what types of backups you are going to do, you
then have to figure out how long you are going to keep the tapes. The most
widely recognized tape retention method is referred to as the Grandfather,
Father, Son method. This system works by using a tape rotation method of full
and partial backups for each day of the week or month.
Once a month, a full backup of the system is done, and the backup is
stored offsite. This tape is never recycled, and it is the tape that is brought out
if all else fails. This is referred to as the Grandfather tape.
The Father tape is the most recent full backup, usually done at the start of
the week. The Sons are the daily backups that contain only the information
that has changed that day. These tapes do not have to be kept forever and
they can be recycled every time a new Father is created and tested. You
should have several weeks worth of Father tapes stored away from the office
in a secure, fireproof, and anti-magnetic environment.

Restoring Data
What is the old saying, The proof is in the pudding? Well, when it comes
down to disaster recovery, the proof is in the restore. Every commercial
backup program gives you the opportunity to verify your data as you are
backing it up. But although every commercial backup program has that
functionality, you still need to check it!

As part of your backup strategy, make sure there is a time when you create
a dummy file, back up the dummy file, delete the dummy file, and restore the
dummy file. If you cannot do all those things, your backup program is not
working.

The differences in the types of backups will also come into play when you
start thinking about restoring data. For example, when I talked about full backups, I said that the full backups take longer than any other kind of backup. That
is true. What is also true is that if the data all has to be replaced, restoring completely from a full backup is the quickest method of getting the server back
online. You only have one tape to mess with, and there is only one pass through
the file system.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

284

Chapter 7

Server Etiquette

If you do not have time to do a full backup every night, you can choose
the method with the next-fastest restoration, which is to do a full backup
with a differential. In this case, you would restore the full backup first, and
then do a restore on the last differential backup. That way, all the files will
be restored to the date and time of the last backup. Although not as fast as
restoring from a full backup, it is faster than the next method, which is the
full backup and the incremental.
With a full backup and incremental, you have to restore the full backup
first, and then restore each of the incremental backups in the proper order.
If you should happen to get things out of order, you may have corrupted the
data. Here is an example. Lets say you suffer a meltdown on Thursday. So
you restore the full backup from Sunday night, the incremental from Monday, and then grab the wrong tape. You restore Wednesday, and then you
restore Tuesday. There are some files that are going to be commonly used
daily, such as the accounting package. These files will be a day old, and
unless the accounting department catches the error, Wednesdays work may
never get added back in. Also, restoring from the combined full and incremental backups takes the most time.
Now that we have covered backups and restores, lets take a look at physical
file server security.

Real World Scenario

If I had to point to one area of the book that is vitally important to your career
as a networking professional, the previous discussions of backup would be
the area. I cannot tell you how many careers have been saved or lost because
of backups. It seems that everyone understands the importance of backups,
but because it is a normal, routine, daily activity, it is often taken for granted.
Backups are taken for granted until one of those "Oh darn" moments (or
something stronger) occurs and you really, really need the tape backup. At
that point, you will know how vital they are.
I have stressed this point above also, but I will stress it again. Be sure to test
the tapes frequently. Doing the backups is wonderful, as long as you can
guarantee that there is really data on the tape. I have lived through a very
painful time when the IT department lost all data for a three-week period
because, although the tape backup software was reporting that it was backing up and verifying the data, in fact it wasnt. It took months to recover all the
lost data, and even then, we could never be sure it was all back.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Physical Server Security

285

Physical Server Security

Back before I was an instructor, I had a real job as a network technician for a systems integrator in Minneapolis/St. Paul. One of the people I
worked with had a weekend upgrade where he almost completely rebuilt the
server: adding memory, hard disk space, and a new tape drive; upgrading the
backup software; patching the system; and more. When he got ready to
leave, he slapped a tape in the internal tape drive and kicked off a full backup
of the new, fully functional system.
Monday morning rolled around, and when he got in, he had several voicemails
from the customer, all complaining that the clients were getting messages that indicated the server was not available. He rushed over to the client site, all the while
thinking, What could I have fouled up? When he arrived, he rushed to the computer room, opened the door, and found out why the server couldnt be located.
Somebody had stolen it. The whole thing! Talk about implementing a disaster
recovery plan.
There are dozens of similar horror stories, but we are here, not to swap tales,
but to talk about how to protect a server by providing physical security. The
whole goal is to make sure that you can provide as much uptime as possible. If
you look at what causes hardware failures, the two primary causes are power
spikes and heat. So, you can go along way toward eliminating hardware failure
by eliminating problems with power spikes and heat. That means, make sure
your server is kept in a room that is air conditioned, or otherwise cooled, 12
months a year. To eliminate static, make sure the room has a humidifier. Finally,
your Uninterruptible Power Supply should also have power conditioning to
even out those highs and lows that come from the power company.
Okay, one more tale from the hardware side. I was doing hardware support
for the local power utility and I got a call from one of the power generating stations that one of their servers had stopped working. I drove to the power station,
and started walking toward the server room. As I was walking toward the server
room, I called my office and told them to send out the full hot spare server by a
delivery service, immediately. They should also dispatch another engineer,
because we were going to have our work cut out for us. The person who was
taking me to the server room asked if I wasnt being a little premature, but I
pointed out that the smell of burned silicon that permeated the building made
any other troubleshooting superfluous. We had them back on line in about two
hours.
What else can you do to protect your server? Put it behind a locked door,
and limit who has access to the server room. Too many companies put the
server behind a locked combination door and then allow everyone in the
company to have access to the combination. Every time the building gets
Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

286

Chapter 7

Server Etiquette

warm, people head to the computer room because it is cool. The computer
room begins to look like the annual company meeting!
Locking up the server room also prevents people from getting physical
access to the servers. Hacking into a system or breaking passwords is very
easy if you have physical access to the system. Keeping unauthorized people
away from the system also prevents many of the oops accidents. These are
the accidents where people hit the wrong switch and turn a computer off by
mistake. Protect the server by putting guards over the on/off switch. If there
is a big power switch in the back, make sure it is secured so cables cannot
wrap around it and turn it off by mistake.
Protection begins with the hardware. Make sure that whatever equipment
you are using for servers is secured away from people, behind a locked door,
in an environmentally controlled, power conditioned room. If you work for
a small office, dont give in to the temptation to save $1,000 by using your
Windows 2000 Server as a workstation also. That $1,000 will be used up
quickly by downtime caused by human error.

Everything you wanted to know about tape, but didnt know

who to ask!
Popular computer tape backup mechanisms use a variety of different types of
tapes. Some acronyms you may see bandied about include QIC, DLT, 8 mm
and others.
In 1972, QIC (Quarter Inch Cartridge) tapes were introduced by 3M. They
were cheap, and just perfect for standalone PCs. The QICs left something
to be desired for networks, because the technology was very slow, and the
capacity was limited. In the mid-1980s, engineers looked at using the
popular 8mm VHS. This was introduced primarily for the UNIX market,
and had a 2.4 GB capacity.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Physical Server Security

287

The next advance came from HP and Sony when they entered the market
using digital data storage (DDS) Technology. DDS used a 4mm metal particle tape similar to tapes used for digital audio recording, known as DAT
tapes. These tapes had a capacity of 1.3 GB on a 60 meter tape. Eventually, DDS tapes grew in size to 125 meters, which allowed for the backup
of 12 GB. This was referred to as DDS-3 and was released in 1995.
In 1996, Exabyte introduced a new technology called Mammoth. Mammoth tapes had a 20 GB capacity with a transfer rate of 3 MB per second.
Mammoth-2 tapes were introduced in 1999. Mammoth-2 adds a new technology called Smart Clean. When the drive is dirty (a very common problem in tape backups) the drive will clean itself without any user
intervention.

Server Room Hints

Here are some other server room hints that you may find helpful.
If you have the opportunity to design your server room, placement can be
very important. The server room is usually given the space that no one else
wants, and in many cases, that is fine. Be aware, however, of environmental
things that may affect performance of your computing environment. One
computer room that I saw had the entire cluster of cables going to a punch
down block running down the middle of the wall. On the other side of the
wall was an elevator shaft. Every time the elevator would go up or down, the
magnetic force created would cause havoc with network communications.
That was a very expensive problem to fix, and it could have been easily
avoided.
If you have been into a computer room, you know that the cables will usually come in from the floors, and be anchored in a punch down block. Then,
there is a series of patch cables that patch the punch down blocks into the
hubs/concentrators/switches or multistation access unit (MSAU). Keep those
punch down boxes turned away from traffic. If the cables are exposed, people
can inadvertently rub against them, loosening the connection.
Color-code your cables. If your server room provides access for multiple
floors, make sure the punch down blocks are segregated by floor, and then
color-code your cable. Document everything you can, from which workspace is serviced by a particular punch down block to where the patch cable
is plugged in. It will help later if you are trying to locate a communication
problem.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

288

Chapter 7

Server Etiquette

Put a telephone and a workstation in the server room. There is nothing

more frustrating than to have a problem that involves calling tech support
and the phone is in another room. Invariably, you will be wasting tons of
time running between the desk where the phone is located and the computer
room getting or giving information. Having a workstation there lets you log
on to the server and test functionality as well as download information from
the Internet.
With all this talk about performance and uptime, what happens when you
have to take a server out of service for a short period of time?

Real World Scenario

Having a lock on the server room door is such a small thing, but it will impact
your life in many ways.
One of the things that I have noticed through my years of working with
networks is how helpful and curious people can be. They are usually at
their best whenever there is a server outage. When I planned server outages, I would plan them days in advance, and would notify all the people
on the network that a particular server or service was not going to be available between these hours on this day. When the day of the outage would
come, I would also make sure that I notified everyone that the server
would be coming down in X minutes, and I would even call stragglers to
get them to exit the server before I shut it down. Once the server came
down, it never failed that people would find me to tell me the server was
down or the network was unavailable. The fact that I was up to my armpits
in file server when they told me didnt seem to make an impact.
For those people who had already scoped out the fact that I knew the network was down, there was the inevitable question, "When will the server
be available?" I would have to answer that question about a hundred times
for every outage. The answer, "When I am finished," did not usually suffice
either. They wanted specifics, like, if you reboot your system at 4:29:15
P.M. and it takes 30 seconds to reboot, you will be the first on the network,
because I am going to bring the server back online at exactly 4:30. It just
doesnt work that way!

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Shut It Down

289

Having a lock on the server room door means they cant get in to bother
you. You can shut your phones and pagers off and really get some work
done!
Seriously, besides the lock, having an alarm system that will notify someone of excessive heat, water, or fire is also an excellent addition to the
server room. If your fire alarm also triggers a fire suppression system,
make sure the entire staff is trained to know what will happen if fire breaks
out. Some of the fire suppression systems do an excellent job of putting
out the fire by removing oxygen from the room, but this is not usually
healthy for your average human being.

Shut It Down

epending on the server operating system that you are familiar with,
you know that in most cases, bringing a server down is not a complicated
procedure. If you are using Windows NT/2000, pressing Control-Alt-Delete
and choosing Shut Down will accomplish the task. In NetWare, it can be as
simple as typing the word Down. But what are the ramifications of downing
a server and how should you prepare?

Downtime
Call it routine maintenance, call it an upgrade, or call it an outage. It doesnt
matter what you call it, the server will be unavailable. Now, this may seem
overly simplistic, but there are two kinds of outages: a planned outage and an
unplanned outage. The planned outage is something that can be scheduled in
advance. The unscheduled outage happens when a server goes down suddenly,
without any warning. In either case, there are users without services. Users
without services tend to be unhappy users.

Unplanned Outage
This is the worst kind. In Windows NT/2000 it can be called the Blue Screen
of Death. NetWare has an even more mystical term, the abend (for abnormal
end). No matter the operating system and no matter the term, it means something has gone wrong.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

290

Chapter 7

Server Etiquette

For a server administrator, the frustrating thing about many of these

issues is that you can never really pinpoint exactly what happened. And
that is the first question everyone asks, What happened to the server?
Sometimes the end comes because of a variety of random acts that just happened to hit at exactly the same instance, causing the server to decide to
stop working. When that happens, shutting the server down as politely as
possible and then restarting it may clear the problem.
In other cases, the cause is easier to identify. You have had a hardware
component go south. Some new patch that you have just applied is not agreeing with the rest of the system and is causing the system to halt unexpectedly.
Fixing the misbehaved component will usually solve the problem. That much
is obvious, though sometimes fixing the problem is easier said than done. For
example, if you have just applied a patch that does not want to work and
play well with other components, keeping the server up long enough to
unapply the patch (if possible) can be your first challenge.

Planned Outages
Things are usually much calmer surrounding a planned outage. You know it is
going to happen, so you can notify the users on your network that the affected
server will be taken down. You can make sure that they know the applications
or services offered by that system will be unavailable, and for approximately
how long. You can also notify your users that they need to exit from the server
or disconnect from the server before you take it down.
You can also make some decisions on the impact the outage will have, and
how you can minimize that impact. For example, taking down a server that
is a member server in a Windows NT 4 domain will have less of an impact
on the overall network than taking down the Primary Domain Controller.
What if the server you are taking down is the Primary Domain Name Service
server for a Windows 2000 Active Directory implementation? How can you
make those services available on another system to minimize the interruption
of service? All of these things are implications that need to be addressed
before you shut the server down.
One of the key questions here is, How long will the server be down?
Now, there are two ways of answering that question. The first way is the
answer you give to your users and to your boss. The second answer is the one
that you really feel. Here is how it works in my world. Say that I am going to
be bringing a server down for an upgrade, and I am going to add more memory
to the system. I know that it may take anywhere up to 15 or 20 minutes to

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Shut It Down

291

politely stop all the services and turn the server off. At that point I can take the
top off, add the memory, put the top back on, turn the server on, make sure
the memory is recognized, and the server will come up on its own volition.
Total timeperhaps an hour. That is what I really think it will take. Now,
what do I tell my users? Four hours at a minimum. Because I know that nothing ever works as smoothly as I think it should. Something will always go
wrong to foul things up a little bitjust enough to throw you off your time
estimate. Now, I know from experience that if I think a job is going to take an
hour, and I tell my users that the job is going to take an hour, and it takes 90
minutes, people think I am an idiot. I may have done everything exactly right
and in record time, but it doesnt matter, I said 60 minutes, it took 90, and
therefore I am an idiot. If, on the other hand, I announced the four-hour minimum, everyone was mentally prepared to be without service for four hours.
Same job, same result, 90 minutes. Now, instead of being 30 minutes late, I am
two and a half hours early. I am a genius. I like being considered a genius much
more than an idiot, so I set expectations where they can be reasonably met.
Before any planned outage, you should make a list of what services the
server provides and see if there is an easy way to duplicate those servers
somewhere else. For example, suppose the server is a DHCP server for a network segment. If you run a traditional 8 to 5 shop, taking the DHCP server
down at 7 P.M. for a couple of hours may not cause undue hardship for your
users. If you are working in a 7-day-a-week, 24-hour-a-day shop, taking a
DHCP server offline could have more severe consequences. That would have
to enter into your plans.
If the server is a Primary Domain Controller or contains the Novell Master
NDS replica, you may want to temporarily transfer those responsibilities to
another server. That way, the network authentication can continue as always,
without an interruption in service.
When you have a planned outage, always plan for the worst. If you try to
cover every contingency and something does go wrong, you will at least have
a plan to solve the problem. In addition, that plan probably will include covering your bases before you downed the server. This can be as simple as making sure you have backed up the server before downing it, or making sure
that you have the proper tools or software on hand if something should go
wrong and you need to get into the box.
There is a polite way to shut down each of the popular network operating
systems. In each case, simply turning off the power is not the preferred way.
All these systems will have files open and working in the background. If you
just dump the power, you can corrupt the file system.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

292

Chapter 7

Server Etiquette

Shutting Down NetWare

There are three ways that you bring down a NetWare Server, depending on
the purpose of the outage. In each case, a command is issued from the
servers command prompt.

Down
When you issue the Down command, the server will stop all the services that
are running and unload all NetWare Loadable Modules (NLMs) that are in
memory. When the services have been stopped and the NLMs unloaded, the
administrator is prompted to enter the Exit command to return the server to
a DOS prompt.

Restart Server
The Restart Server command performs all the tasks that bring the server down,
and then will simply restart it again. If there are no command line switches
employed, Server.EXE will be reloaded using all the appropriate NetWare
Control Files (NCF). This is useful if you are working on the configuration of
the server and you need to reload the operating system. In addition, Restart
Server can be run with several parameters to enhance its functionality. These
include:


-NS. If you use this parameter, Server.EXE will load but it will not
use the Startup.NCF file.

-NA. If you use this parameter, Server.EXE will load but it will not
use the Autoexec.NCF file

-d. If you use this parameter, you will break into the internal debugger.

Reset Server
The Reset Server command, also issued at the command line, brings the
server down by stopping all the services that are running, unloads all the
NLMs, and causes the server to perform a warm boot. If Server.EXE is in
the autoexec.bat file, it will execute and the NetWare server will come
back online.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Shut It Down

293

Shutting Down Windows NT/2000

Shutting down a Windows NT/2000 server is performed either from the
Start menu by choosing Shut Down, or by using the Ctrl-Alt-Delete keys to
open the Windows Security window and then choosing Shut Down. Like
NetWare, Windows gives you the opportunity to shut down the server in
expectation of turning it off, or to shut it down and immediately restart it.
Restarting the server shuts down all the services and performs a cold boot.

If your Windows NT/2000 server is running Exchange, you may want to go

into Control Panel, Services, and manually shut down Exchange. This seems
to speed up the shutdown process. With Exchange running, the process can
seem to take forever!

Shutting Down Unix/Linux

When talking about Unix and Linux, you are not talking about one specific
operating system. There are many different versions. Some may use some
advanced features and some may not. Like Windows and NetWare, you just
cant turn off the power switch on a Unix/Linux system. Like NetWare, there is
a command to close all services and prepare the machine for being powered off:


shutdown [minutes] [warning]

Minutes will specify the number of minutes the system will wait before
performing the shutdown. Warning is an optional message that will be displayed for any users that are logged in. Some versions will allow you to use
the word now as part of the shutdown command. In addition, you can add
-r to the command to force the system to reboot.
If you want to shut the system down immediately, you can also use the
Halt command, or perform Ctrl-Alt-Delete. In either case, the system will be
stopped immediately and the power can be shut off. The three-finger
salute (Ctrl-Alt-Delete) is not universally recognized across all versions of
Unix and Linux.

Shutting Down OS/2

OS/2 is another 32-bit operating system that requires all services to be shut
down before the system can be powered off. With OS/2, shutdown is accomplished using the Ctrl-Alt-Delete key sequence.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

294

Chapter 7

Server Etiquette

Real World Scenario

It is the little things that make a difference. This is especially true when it
comes to how people perceive you and the way you do your job. Once you
start getting your hands dirty working with servers, you will learn that
there are several ways you can alter the way people feel about you. You
may be the brightest tech in the world and know just about everything
there is to know, but if the customer doesnt believe that, you are going to
have a rough time. The worst part of this scenario is that you cant tell the
customer you are competent, because they wont believe you!
Here is what I mean. If you walk into a server room and start to do something
to the server, and it reacts in a way you didnt expect, saying something like,
"Darn, I never saw one of these do that before! I wonder why that happened?"
will probably not endear you to the customer. Keeping your mouth shut and
acting like you have seen it before (while experiencing a serious panic attack
on the inside) will lend credence to the fact that you know what you are doing.
I had one case where I was building a server, and the salesperson told the
client that it was perfectly acceptable to come in and watch the technician
configure the new file server. I just loved it when that happened! When I
took the server out of the box, it was in the form of a cube, and I needed
to get inside the server to add components. I had my trusty screwdriver in
hand, and I began looking for the screws that would gain access to the
innards. There werent any. Anywhere. I turned the server over and over,
looked at the top, the bottom, and all the sides and there were no screws,
no magic buttons to push, no removable sides, nothing that I could see. I
literally had to dig out the manual to figure out how to get the top off this
darn server. Then, it turned out the front and sides were held in place by
Velcro. When I tried to pull the front of the box off, the plastic started to
bow, looking like it was going to break. It didnt, but still, the customer
wasnt really impressed with the fact that the tech who was building their
server had to read the instructions to figure out how to open the darn thing
up. It really didnt do much for my credibility.
Some things are best done away from the prying eyes of the customer.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Summary

295

Summary

There were lots of different topics covered in this chapter, werent

there? There was everything from the different types of application servers to
how to stop a server. When you look over the material, do it carefully,
because there are lots of salient points that can be questioned.
In the next chapter we will look at ways to keep the footprint of our servers
small by using rack mounts. Rack mounting servers means more than just
installing the system in the rack. There is planning that needs to be done so
cables are not strewn all over the room and so that you can have multiple computers using one monitor, one keyboard, and one mouse.

Exam Essentials
Know what a dedicated application server can provide Dedicated
application servers provide application services in a client/server environment. This can be something like a dedicated Web server or a mail server.
Know when to use a distributed application server A distributed
application server would be something like a database server where
some of the data is stored at a remote site.
Know how a peer to peer application server is used An example of a
peer to peer application server would be in a gaming environment, where
several users are playing the same game over the Internet.
Know the different types of backup strategies Know what each backup
type accomplishes. Be sure to review the Grandfather, Father, Son method
of tape rotation.
Know how to physically secure a server and server room Server rooms
should be secured with a locking mechanism on the door, at the very least.
This can be a key lock, a combination lock, a pass-card type system, or
even a biometric system. Access to the server room should be limited, and
frequent security audits of the room should be completed.
Know how to shut down a Windows NT/2000 Server The keystrokes
necessary to shut down a Windows NT/2000 Server are Alt-Ctrl-Delete
and choose Shut Down, or choose Shut Down from the Start menu.
know how to shut down a NetWare Server To shut down a NetWare
Server, you type down and press Enter at the server prompt.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

296

Chapter 7

Server Etiquette

Know how to shut down an OS/2 Server The keystrokes necessary to

shut down an OS/2 Server are Alt-Ctrl-Delete and choose Shut Down.
Know how to shut down a Unix Server UNIX and Linux Servers are
brought down by issuing the shutdown command.

Key Terms
Before you take the exam, be certain you are familiar with the following terms:
abend
Copy backup
Ctrl-Alt-Delete
dedicated application server
Differential backup
Disaster recovery
distributed application server
Down command
Full backup
Incremental backup
level of risk
multistation access unit (MSAU)
peer to peer application server
planned outage
Reset Server
Restart Server
Shut Down
Uninterruptible Power Supply (UPS)
unplanned outage

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

297

Review Questions
1. Why is it important to perform a graceful shutdown whenever possible?
A. To avoid file corruption
B. To prevent electric charge to electronic equipment
C. To gracefully spin down hard drives
D. To gracefully disconnect users from NOS
2. Ming has a server that has been acting very peculiar latelynot at all

like its old self. What steps can Ming take to diagnose the problem?
A. Perform hardware diagnostics.
B. Reboot the server.
C. Read the event logs.
D. Replace the hard drives.
3. Moshebi has a network with many wake-on-LAN adapters in it. One

of the adapters fails to wake up and find its boot server and the computer wont come up. How can Moshebi troubleshoot the problem?
A. Read the servers event logs.
B. Read the workstations event logs.
C. Ping the adapter.
D. Check the adapters BIOS and configuration parameters.
4. Name some field replaceable units (FRUs) that you could easily

replace on a server.
A. Power supply
B. Cooling fan
C. Case
D. Drive cage

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

298

Chapter 7

Server Etiquette

5. Nell is getting an alert from a server in another office several hundred

miles from her company headquarters location. The alert is telling her
that she has a battery that has failed in the servers UPS. What are
some things that she can do from her location?
A. Telnet into the UPS to reboot it.
B. Issue a shutdown r command on the server.
C. Ask someone in the remote office to get a repairperson to replace

the battery.
D. Review the error logs.
6. Which backup strategy will take the longest amount of time to back

up and the least amount of time to restore?

A. Full
B. Partial
C. Incremental
D. Differential
E. Copy
7. Which backup strategy will not reset the archive attribute on a file?
A. Full
B. Partial
C. Incremental
D. Differential
8. Which backup strategy backs up all files that have been changed since

the last backup?

A. Full
B. Partial
C. Incremental
D. Differential
E. Copy

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

299

9. Risk assessment is part of what process?

A. Backup and restore
B. Disaster recovery planning
C. Network security
D. Group Policy Objectives
10. Name some places you might check to get tips or fixes for a problem

youre having.
A. Newsgroups
B. Intranet
C. Internet
D. Knowledgebase CDs
11. Mary Beth is a new server admin in a large corporate environment.

She knows how to restart servers, read logs, add and remove users,
and change passwords. What are some tasks that she should seek
senior administrator assistance with?
A. Logging on to the server
B. Stopping and starting a daemon/service
C. Windows 2000
D. Working on a cluster server
E. Resetting a fax board
12. When do you need to terminate an externally connected SCSI device?
A. When its on the end of the chain
B. Always
C. Only on devices that arent self-terminating
D. Never

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

300

Chapter 7

Server Etiquette

13. When mounting a rack installation, what are some electrical

considerations?
A. Provide proper grounding of rack.
B. Never use the electrical strip that comes with racks.
C. Make sure voltage and amperage requirements are met.
D. Always use the electrical strip that comes with racks.
14. Select the external devices that you could replace, while the server was

online and running, without creating a problem with most network

operating systems.
A. Backup tape drives
B. Keyboards
C. Mice
D. External disk arrays
E. Modems
15. Which backup and restore strategy will take the least time to back up,

but the longest time to restore?

A. Full backup
B. Incremental Backup
C. Differential Backup
D. Copy
16. Reva is a network administrator whos rebuilding an older server. As

she boots the computer everything looks OK, but when she begins to
install the NOS software she notices that the floppy disk drive no
longer works. What signs could she look for at POST?
A. BIOS error reporting floppy failure
B. Beeping from motherboard
C. LEDs operating on floppy
D. Floppy firmware reporting an error

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

301

17. What are some considerations to think about when purchasing a

small UPS?
A. Total amperage rating
B. Number of available sockets
C. Upgrade abilityWhether a serial cable can be used to connect to a

server
D. Whether a serial cable can be used to connect to a server
18. Lucy is purchasing some servers that can connect to the switch on the

network with fiber optic cables. What options are available to Lucy
when considering supplying the fiber optic cables that run from the
servers to the switch?
A. Purchase already terminated cable on spools.
B. Purchase terminated cables manufactured to designated length.
C. Retain a cabling specialist to create exact-length cables.
D. Create the cables herself.
19. What is not a type of backup strategy?
A. Full
B. Detrimental
C. Differential
D. Incremental
E. Copy
20. Why is it not a good idea to leave a servers cover unscrewed and

merely hanging on the chassis?

A. Provides a mechanism for electricity to accidentally be applied to

the chassis.
B. Dust and moisture can get to internal parts.
C. Someone might trip over open sides.
D. Power supply wont work until cover is screwed on.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

302

Chapter 7

Server Etiquette

Answers to Review Questions

1. A, D. Just callously hitting the power button can and will corrupt files

and, if youre like me, youll be lucky enough to have some of the files
that are corrupted be files that are critical to the operating systems
functionality! Powering down the computer before first downing the
NOS wont introduce a surge to the computer, but you do need a surge
protector. In either case, as soon as you power down the computer, the
drives spin down. By performing a graceful shutdown you may be able
to provide time to your users to log off before you shut down. Some
NOS packages automatically inform the user the servers going down,
and others require that you manually intervene, but this is a hugely
important feature of a graceful shutdown.
2. A, B, C. Replacing the hard drives without first garnering more

information is jumping the gun. You can and should perform some
hardware diagnostics. You should try rebooting the server. Some
servers seem to "feel" better (forgive the anthropomorphism) if
theyre regularly booted. (Thats because some applications arent
written very well and dont recapture memory thats no longer in
use. Booting frees up memory thats being held but shouldnt be and
sets the system right again.) Reading the event logs might provide
key diagnostic information.
3. A, C, D. Reading the servers event logs might provide some clues as

to why the NIC isnt connecting. You can also ping the adapter. The
OS on the workstation isnt booted yet, so you cant check the event
logs to see if you can find out whats wrong there. The adapters BIOS
and config parameters might reveal a misconfiguration of some kind.
4. A, B, D. An FRU is something thats designed to be easily and quickly

replaced. The power supply, cooling fans, and drive cages are typically
designed in such a way. If you have to replace the case, youre talking
about a fairly extensive operation entailing pulling the motherboard,
disconnecting all cables, etc. Though its field replaceable, its certainly
not a quick and dirty chore.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Answers to Review Questions

303

5. C. This is a case where a remote alert doesnt do Nell any good

because she cant do anything about the problem. Perhaps cycling the
UPS or server might clear up the problembut its doubtful that it will
do so. Reviewing the error logs would reveal the same information as
the alert. Shes stuck with asking someone in the remote office to have
a repairperson come in and check it out.
6. A. A full backup will take the most amount of time to back up the

files, but the least amount of time to restore the files.

7. D. A differential backup will not reset the archive attribute on a file.
8. C. An incremental backup will back up only those files that have been

changed since the last backup.

9. B. Risk assessment is part of the disaster recovery planning process.
10. A, C, D. Options A, C, and D are all common places to check for help

with a problem and/or downloading of a patch. You probably wont

have much help on the intranet unless theres a knowledgebase on it
that has been specifically designed for such a task.
11. C, D. If theres proprietary hardware, such as a fax board or clus-

tered server, Mary Beth should first obtain some training from senior
administrators or even a class or two on the subject, then attempt
to administer it.
12. A. The first and last devices on the chain need to be terminated. The

SCSI card is always at ID 6 and, in todays modern architectures, selfterminates. The other end of the chain, whether thats ID 0, 1, 2, or
whatever, needs to be terminated. If the external device represents the
opposite end of the chain, then youll need to terminate it.
13. A, C. Some racks come with electrical strips mounted and ready to plug

in. If you intend to use them, its important to make sure that the socket
youre using has a circuit with a high enough amperage to handle all of
the devices you might plug into the strip and that its properly grounded.
You should also consider grounding the rack itself. Whether you use the
power strips that come with the racks is up to you.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

304

Chapter 7

Server Etiquette

14. B, C, E. As a very general rule of thumb, youll find that backup tape

drives and external disk arrays arent usually recognized if theyre

taken off of the server and replaced with a new unit or simply put back
on. Keyboards and mice dont usually have this problem. Note that
this varies with the NOS and its version number, and with the device
drivers for the various components. But its not a good idea to remove
any device while the machine is powered up.
15. B. An incremental backup wont take as much time to back up, but

you will have restore each backup session in order.

16. A, C. Generally the BIOS will report floppy disk failures, although

the way that the failure is reported varies from BIOS to BIOS. Youll
usually not hear beeping from the motherboard on floppy disk drive
failures. You can also watch to see if the LEDs operate on the floppy
drive. Thats a sure sign that its getting power and that the power
cable isnt reversed (quite commonthe LED will be solid on if thats
the case). While floppy drives have firmware, its not smart enough to
report that it has somehow failed.
17. A, B, D. As a general rule of thumb, theres very little to upgrade in

a UPS. You can replace batteries and upgrade the server software that
monitors them, but thats about all. The amperage rating (in volt
amperes) will be important because during a blackout the device has
to carry the load of the devices plugged into it. Also you may want to
connect a serial cable from the UPS to the server and monitor it with
software provided by the manufacturer.
18. A, B, C. Unless Lucy is certified to manufacture her own fiber optic

cables, this is one job best left to the pros. She could purchase spools of
fiber optic cabling that have been terminated on both ends. These are
Plug and Play, but buying the excess length could be expensive. You can
also buy predefined fiber optic cable lengths (such as 10 meter) that are
already terminated; his is a more cost-effective method. And, if you have
a lot of cable to get, you may want to consider a cabling expert to make
them to-length for you. Dont forget multimode versus single mode
cabling and Type SC versus Type ST terminations.
19. B. Detrimental is not a type of backup strategy.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Answers to Review Questions

305

20. A, B, C. Especially if a server such as this is in a rack, it would be very

easy for a power surge to be accidentally applied to the chassis and

potentially hose up a lot of expensive electronics. Dust and moisture
are not friends to electrostatic devicesits best to keep particulates
out of the works. And its even possible that a klutz might trip on the
side of a server, potentially knocking it over along with himself.
Depending on the computer, the power supply will usually work fine
with the cover off.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Chapter

Inside the Server Room

SERVER+ EXAM OBJECTIVES COVERED IN
THIS CHAPTER:

1.2 Install hardware using ESD best practices (boards, drives,
processors, memory, internal cable, etc.).


Mount the rack installation.

Cut and crimp network cabling.

Install UPS.

Verify SCSI ID configuration and termination.

Install external devices (e.g., keyboards, monitors,

subsystems, modem rack, etc.).

Verify power-on via power-on sequence.

2.1 Check/upgrade BIOS/firmware levels (system board, RAID,

controller, hard drive, etc.).

2.4 Configure external peripherals (UPS, external drive
subsystems, etc.).

2.9 Document the configuration.

3.2 Add processors.


On single processor upgrade, verify compatibility.

Verify N 1 stepping.

Verify speed and cache matching.

Perform BIOS upgrade.

Perform OS upgrade to support multiprocessors.

Perform upgrade checklist including: locate and obtain latest

test drivers, OS updates, software, etc.; review FAQs,
instructions, facts and issues; test and pilot; schedule
downtime; implement using ESD best practices; confirm that
the upgrade has been recognized; review and baseline;
document the upgrade.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com


3.5 Upgrade BIOS/firmware.



Perform upgrade checklist including: locate and obtain latest

test drivers, OS updates, software, etc.; review FAQs,
instructions, facts and issues; test and pilot; schedule
downtime; implement using ESD best practices; confirm that
the upgrade has been recognized; review and baseline;
document the upgrade.

3.10 Upgrade UPS.



Perform upgrade checklist including: locate and obtain latest

test drivers, OS updates, software, etc.; review FAQs,
instructions, facts and issues; test and pilot; schedule
downtime; implement using ESD best practices; confirm that
the upgrade has been recognized; review and baseline;
document the upgrade.

4.4 Perform physical housekeeping.

5.1 Recognize and report on physical security issues.


Limit access to server room and backup tapes.

Ensure physical locks exist on doors.

Establish anti-theft devices for hardware (lock server racks).

5.2 Recognize and report on server room environmental issues

(temperature, humidity/ESD/power surges, back-up
generator/fire suppression/flood considerations).

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

n the last chapter we covered some of the things you should know
about designing a server room, but in this section we are going to probe a little deeper. We are going to look at things like how to manage all those
needed servers without building a separate building. We are also going to
look at when you will have to apply BIOS and Firmware upgrades, and cover
the topic that everyone knows about, but very few do anything about: documentation. The first part of this chapter is going to be kind of fun, because
I always love spending other peoples money!

For complete coverage of objective 1.2, please also see Chapters 1, 6, and
7. For complete coverage of objectives 2.1, 3.5, and 4.4, please also see
Chapter 6. For complete coverage of objective 3.2, please also see Chapters 3, 6, and 9. For complete coverage of objective 5.1, please also see
Chapter 12.

Rack-Mounted Servers

In the last chapter, we did spend a little time taking a look at the way
the server room should be configured. Lets revisit some of those issues, and
also take a look at why we buy servers instead of just computers. After all,
servers are just PCs with an attitude, right?
Actually most server products are optimized to make sure they have fast
data transfer rates and fast hard drives, as well as more processors and memory than what is usually available in PCs. In addition, one of the considerations you have with a server that you may not have with a workstation PC
is scalability. You may not need to have five drive bays in a workstation, but
it may come in handy in a server.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

310

Chapter 8

Inside the Server Room

Now, the problem with these monster machines is their footprint. Footprint is another of those terms you hear thrown around a lot, and everyone
just expects you to know what it means. Basically, it is just how much room
this thing takes up in your server room. As you may have noticed over the
last several chapters, networks have evolved. It used to be that a department
or company had a couple of file and print servers stuck away in a room
somewhere, but now companies commonly have dozens of servers in a single
location.
Having that many servers can lead to multiple problems, not the least of
which is where do you put them. The footprint taken up by 12 servers, with
monitors, keyboards, and mice, can be considerable. That footprint can also
be expensive. Lets say for the sake of argument that the average server takes
up 3 square feet of floor space. Now, if we are putting multiple servers in the
same room, you want to make sure they are well ventilated, so you are going
to leave at least a couple of feet between servers. If you add in the room that
the monitors, keyboards, and mice are going to take up, as well as the bakers
racks, or tables, or shelves, or whatever you have that you are going to store
these on, it is going to take up some serious amounts of real estate. Now, I
have no idea what office space rental is in your area, but if you multiply the
number of square feet all these servers are going to take up, as well as all the
number of square feet for desk space, wiring cabinets, hubs/concentrators/
MSAUs, uninterruptible power supplies, air conditioning units, etc., and
multiply that times the average cost of a square foot of office space, you are
going to find the cost staggering.
Computer server manufacturers decided many years ago to use the same
philosophy used by people who design buildings for downtown areas. It is
better to go up than out. Now, computer manufacturers will have the traditional tower-based or modular server unit, as well as the units that get
mounted in racks.

Going Up
To start this discussion, lets take a look at what the servers are going to be
mounted into, the actual racks. Features, as well as prices, may vary, and I
am not here to recommend any particular manufacturer. As a matter of fact,
one criterion for the graphic in Figure 8.1 was that it was in line art format
rather than a photograph.
Take a look at Figure 8.1. This is a picture of your ordinary, everyday,
empty rack, which I found on US Logics Web site (http://
www.uslogic.com/rack/code/rack7500.html). At least that is the way it
looks at first blush.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Rack-Mounted Servers

FIGURE 8.1

311

US Logic 7500 rack enclosure

1

6
2
5
3

This drawing shows the following:

1. Top mounted fan
2. Removable casters with wheel locks
3. Adjustable, leveling feet
4. Retractable anti-tilt stabilizer
5. Plexiglas locking front door
6. Vented bottom panel
7. Removable side doors

So, other than a footprint, why use racks? First of all, by combining all
your servers in one location, you are providing ease of access. If I want
to check out your server environment, I can go to the rack, open the
door, and have access to all the servers in one relatively small, easy to
manage location. Because these racks can be hooked together (see the
removable side doors above), they offer an almost infinite range of

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

312

Chapter 8

Inside the Server Room

expansion. With the ventilation at the top and bottom, an open rear
panel, and the optional fan package, a rack can improve the movement of air to dissipate the massive amounts of heat some of these
computers will generate. Finally, depending on the way you configure
your keyboard and mouse, you may be able to secure the entire rack
of servers by locking the front of the cabinet.

Racks will normally have an interior dimension of 19 inches. The company will
specify the height and width of the rack, but what the IT person is mainly concerned with is the usable space, measured in Us. For example, this rack
comes in the Mini-Cabinet of 23U, the Medium Cabinet of 31U, and the Tall
Cabinet of 40U. The U is simply the cutout on the side of the rack-mounted
appliance where the screw fits in to hold the appliance to the rack. Appliances
are then measured by the number of holes they cover up. For example, servers will be listed at 1U, 2U, or 3U. A component that holds 5.25-inch hard
drives may be listed at 4U, which means when you install it into a rack, you
will be covering 4 holes. A little piece of trivia for you: a U measures 1.75
inches.

You should take care when placing the rack in your server room. Once the
rack is in place, and all the equipment has been installed, it may not be easy to
relocate. In our sample rack, there were removable casters with wheel locks. If
you were using that type of rack, you could simply unplug everything, unlock
the wheels, and roll the rack carefully to its new location. If your rack came
without wheels, relocation would involve removing everything from the rack,
repositioning the rack, and then putting everything back.
Speaking of placement, there are no real guidelines on what goes where in
the rack. Common sense dictates that the heavy items, like the UPS, should
be placed at the bottom of the rack. The keyboard and monitor placement
will depend on the average height of the operator. Most of the racks I have
seen are designed to use while standing in front of the server console. There
is nothing to prohibit you from configuring the rack so you can sit instead of
standing.
What about the server? How much functionality can a server offer that
fits in a space so small? In this case, look at Figure 8.2.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Rack-Mounted Servers

FIGURE 8.2

313

Compaq ProLiant DL380

1

6
5
4

Again, I am not pushing any line of serverI am using this drawing

because it is line art instead of a photograph. This information was taken
from the Compaq Web site at http://www5.Compaq.com/products/
quickspecs/10495_div/10495_div.html. This server is referred to as
coming in a rack-ready form factor that is 3U, 5.25-inch. This is what is
packed into that 3U form factor:
1. IDE CD-ROM
2. 1.44MB disk drive
3. Protected power switch
4. Front panel status LEDs
5. One full-height or two half-height removable media bays
6. Four 1-inch-wide Ultra2/Ultra3 SCSI hot plug drive bays for up to

145.6GB internal hot plug capability

In addition, this server can handle two Intel Pentium III processors. It will
max out at 4GB of memory. It has four PCI expansion slots, an embedded
10/100 Fast Ethernet NIC, 275-watt power supply, and simplified system
board removal for ease of servicing. These servers can pack a lot into a small
package.
What about other types of things that can be put into the rack? Certainly
there are all types of servers, there are UPS systems, and there are units that
house a keyboard and LCD video display all in a 1U configuration. There are

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

314

Chapter 8

Inside the Server Room

tape units and there are disk storage subsystems. If you dont want the integrated keyboard/LCD solutions, there are plug-in cabinets that will hold a
standard monitor. If you need it, it has been built for rack mounting.

All this equipment tightly backed into one compact efficient package is still
going to generate lots of heat. Make sure your computer room has adequate
cooling and ventilation to maintain the room temperature around 70 degrees.
Heat is really a killer on hardware, and sometimes a temperature rated safe
really isnt.

Rack Security
One of the disadvantages to using rack-mounted devices in your computer
room is also one of the advantages. It is very nice to have complete access to
every server in the room from one area, but from a security point of view,
this can be less than desirable. If you have opted to put your servers in racks,
be sure to take every step necessary to protect the rack. Later in this section,
we will talk about protecting the devices from threats like electrical spikes or
outages, but also protect the rack against physical threats. Be sure to lock the
computer room. The security of the computer room will depend on how
deep your pockets are, but if you can have the computer room accessed only
with a key card, that will help. With key cards, traditionally, when the card
is swiped to gain access, an entry is made in a log that the room was accessed
at that time by that individual.
Racks will usually have doors available. These glass doors are lockable,
and it is a great idea to make use of them. You may think they are a pain
when you are trying to get to the server and you have forgotten your key, but
they can be a deterrent from unauthorized access of the servers. This is an
especially good idea if you share your server room with the phone people or
the Wide Area Network people. Sometimes people get curious and just want
to lookyou know the type.
The more security that you can provide between the keyboard and the
user, the better.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Cable Management

315

Real World Scenario

When working with rack-mounted equipment, plan the placement of
devices to be as easy to access as possible. In other words, make sure the
CD-ROM devices, the DVD devices, the tape backup unit, the monitor, and
the keyboard are all within easy reach. I have come across rack devices
where planning did not play an important part in the process, and it was
difficult to get to some of the most commonly used peripherals.
Make sure your rack device does have a locking door and that it is used. Yes,
it can be a pain in the neck to discover the key for the rack is back at your desk,
but it will pay off in the long run. It is just another way of keeping folks from
messing with your servers. Also, common sense says to make sure things like
full coffee cups, opened bottles of Mountain Dew, and other items stay far
away from the rack enclosure. Spilling coffee at your desk is an inconvenience;
spilling coffee into a rack of computer components is a disaster.

Cable Management

By utilizing a rack-mounted server system, you have an efficient area to

store all your servers and appliances, but think of the cable mess! Figure that
with a 40U rack, if you fill that thing up with devices that average 3U in size,
you could have 13 different devices in that one rack. Cable management in a
situation like that is imperative!
When I think of cable management, my first impression is about ways to
run cables from workstations to the server room to keep them in some semblance of order. They should be in order so that they can be traced and
replaced if necessary. This also makes it possible for your bundles of cables
to be kept neatly out of the way! Now, that is definitely one part of cable
management. But another part of cable management is just making sure that
you have the right cables on hand to do the right job.
Earlier, we used the example of the rack-mounted server. If you were to
check out the cable management portion of the Web site for that server, you
would see page after page of information on cables that you would have to
use, depending on the configuration of the server. For example, there are special cables if you are using Fibre Channel and other special cables for certain
SCSI devices.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

316

Chapter 8

Inside the Server Room

Sometimes You Just Have to Be Lucky

There are times in your life as a network administrator that it pays to get
really lucky. I had one of those experiences. I had a network segment that
suffered from sporadic communications problems. These problems would
hit an entire row of computers, last anywhere from 5 to 10 minutes, and
then disappear for weeks on end. When it hit, everyone suffered. When it
went away, everyone went back to being just fine immediately. The 5- to 10minute time frame meant that by the time they contacted me, and I got to
the floor, the problem was gone. I, of course, told the users that the network
knew I had the ability to unplug it, and therefore it would self-heal as soon
as I showed up. They were convinced; I wasnt.
One afternoon I was leaning on a desk, just killing time talking to a friend of
mine. As I rested on her desk, I was looking down the long hallway at the
janitors closet. While I was watching, the janitor opened the door and
turned the light on, and suddenly everyone started to complain. As I turned
around and started walking to one of the desks, I noticed the janitor shut the
light off. Almost immediately people started to complain that I had done it
again and things were working again. I told them I was going to try to work
some magic. I got the janitor and we opened the door and turned the light
on: Things went south. Light off: Things were good.
Now, it didnt take this boy long to figure out that we had some sort of causeand-effect relationship here. I may not be the brightest bulb on the tree (no pun
intended) but I didnt need a house to fall on me either. I got a ladder, lifted the
ceiling tile, and sure enough, there was the cable connecting the server room
to their cable closet lying directly over the fluorescent light. I grabbed it and
moved it away from the light fixture and the problem went away.
Like I said, sometimes you dont have to be good, just lucky!

Cable management means different things, depending on the circumstance. It can mean having the proper cable to do a particular job, such as
hook a rack-mounted server up to a Fibre Channel Storage Area Network.
Cable management can mean threading the power cable, the mouse cable,
and the video cable through the holes in the back of the bosss desk when setting up her computer. This way it reduces the clutter in her office and makes
it neater, making her happier.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Cable Management

317

For the network as a whole, cable management starts even before the wiring begins: It starts in the design phase of the project.

When you are designing a computer room, or a large network cable run, too
often management figures that we will just let the IT team handle the cabling.
About the time they start talking that way, you should probably mention that all
cabling has to be done in accordance with local and state codes. Some of these
codes may match national standards, and some may be unique to the local
municipality. This is a very good time to look at hiring a professional cabling
firm. They will know the requirements and how to meet or exceed them.

Backbone Cables
The object of the exercise is to move data from the local workstation into the
wiring closet and on into the server room. There are some different classifications of cable runs that make this possible. These are referred to as horizontal cable, backbone (vertical) cable, and patch cables.
Now, you may think that horizontal cable and vertical cable refer to their
reference to the horizon, but that is not the case. Cabling people have their
own language too! The cable that runs between the panel in the wiring closet
and the wall jack is referred to as the horizontal cable. The backbone or vertical cable is the cable that runs between the wiring closets and the main connecting point of the building. This could be called the equipment room, or in
terms of networking, it may be called the server room.
In terms of types of cable, if you are using Ethernet or Token Ring, the
horizontal cables are probably 100-ohm, four-pair unshielded twisted
pair (UTP). You may commonly hear this referred to as Cat 5 cabling, for
Category 5. There are some other types of cabling that can be used for
horizontal runs. For example, you can use shielded twisted pair (STP),
multimode optical fiber.
The same types of cables can be used as the backbone. While you can use
UTP or STP, the attenuation properties of these make long runs between
floors or between areas impractical. That is why many backbones use fiber
optic cables as the preferred media. Besides providing the ability to successfully traverse longer runs, fiber optics is not subject to Electromagnetic Interference (EMI), which means it will provide a cleaner signal.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

318

Chapter 8

Inside the Server Room

For a much more thorough coverage of cabling, see Cabling: The Complete
Guide to Network Wiring, by David Groth and Jim McBee, Sybex, 2000.

Cable Paths
Since we have to get these cables from the work area to the server room,
someone is going to be playing around inside the ceilings and inside the
walls. Any time you start running wires in the walls and ceilings, you have
to pay attention to code. So, lets look at some of the cabling components
that the Telecommunications Industry Association (TIA)/Electronics Industry Association (EIA) 569-A Commercial Building Telecommunications
Pathways and Spaces Standard lays out. I will be taking a look at some of the
pieces that are used in work areas and wiring closets, as well as horizontal
and backbone cable runs.

Conduit
Conduit can be either metallic or nonmetallic. It can be rigid or it can be flexible
piping, depending on your local code. Conduit will run from the work area to
the wiring closet or server room.
Conduit has several advantages. The major advantage is that it may be
already present in the building. If (and this can be a big if) there is space left
in the conduit, it should be easy to pull cables through it. The problem with
conduit is that it offers a limited amount of space and once that space is filled
you are stuck with coming up with another alternative.
According to specifications, conduit can be used to route either horizontal
or backbone cables. If you are working in a multistoried building, you can
use fire-stopped conduit to link floors.
Check your building codes; some localities require conduit for all cable,
either network or electrical.

Cable Trays
Take a look at Figure 8.3. This is an example of how cable trays can be used
successfully. You will notice that this implementation is using two different
styles of cable management: the J-hook approach and the ladder rack.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Cable Management

FIGURE 8.3

319

Cable runway system to suspend cables overhead

J-hooks

Ladder racks for managing

cables in equipment room,
in telecommunications closet,
or in the plenum

19" rack

These can both be considered types of cable trays. They are designed to
hold the weight of all the cables, and provide an ideal way for handling a
large number of horizontal runs. You simply lay the cables inside the tray.
That way, the cables are easily accessible for either maintenance or for troubleshooting.
If you look at Figure 8.3 carefully, you can see why each type got its name.
The J-hooks are just hooks that are anchored to walls, and the cable bundles
are laid in the crook. The ladder rack resembles the extension ladder you
may have in your garage. It can be mounted either in the space between the
ceiling and the ceiling tiles (this is called the plenum) or it can be exposed in
the equipment room. The vertical ladder can also have D-brackets installed
on the rungs to keep the cables from getting unruly.
There are other advantages of the cable tray system. They keep cables
from being draped over the ever-present fluorescent light fixtures or the
heating and air-conditioning equipment. When this system is used in the wiring closet or server room, it keeps the cables off the floor or can direct them
to the patch panel or equipment racks.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

320

Chapter 8

Inside the Server Room

Raceways
A raceway is another one of those things that I knew existed, and I knew
what it did, but I was never sure of the name. I always associate raceway with
my friends in Indianapolis!
Raceway is the name given to the conduit that is used for surface mounting cables on walls or in cubes. Take a look at Figure 8.4 to see an example
of a surface-mounted raceway system.
FIGURE 8.4

Surface-mounted raceway

All of these systems are designed to keep the cable runs out of sight, off the
floor, and away from things that can cause problems. The things that can
cause problems range from the seen, like fluorescent lights, to the unseen,
like unclean power.

Electrical Protection
Sometimes we forget that all these cables that we are managing carry electrical
current. Besides providing physical protection against accidents, such as people pulling out plugs or running over a fiber cable with heavy equipment, you
should also provide for electrical protection. Variations in power can cause all
sorts of problems, ranging from having to reboot a computer after a short outage to sustaining component damage.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Cable Management

321

I once attended a discussion of clean spike-free power. Someone once commented that it seemed odd that a component that could make it through a
severe thunderstorm would suddenly fail for no apparent reason. The person
giving the lecture said that the component is like a heavyweight fighter who
gets knocked out. Usually, the knockout punch is not the first punch thrown,
and in many cases, is not as powerful a punch as the fighter had absorbed
before. It was the cumulative effect of all the punches that caused the fighter to
go down. The same is true of computer components. They may take dozens of
hits in their lifetimes, and each causes small amounts of damage. When the end
comes, it is the cumulative effect that causes the failure.

There are several different levels of protection available in the marketplace.

We will look at a standby power supply (SBS) and an uninterruptible power
supply (UPS)!

SBS
The SBS contains a battery, a switchover circuit, and a device that changes
the DC voltage from the battery into the AC power that your servers will
demand. This device is called an inverter.
The outlets on the SBS are all connected to the switching circuit. The
switching circuit is connected to the buildings AC power. The switching circuit keeps an eye on the incoming AC power line voltage, and if it drops
below a preset limit, the SBS switches to battery power. When the voltage
comes back in range, the SBS will switch back to AC. Some SBS units not
only can control the lower-than-expected power, but can also be on the lookout for power spikes. If the voltage goes too high, the SBS will kick in before
the spike reaches dangerous levels. This is referred to as power conditioning.

UPS
As the name implies, an uninterruptible power supply is another type of battery
backup. It is different from the SBS in that the equipment is always running off
the inverter and the battery. The UPS is constantly charging and monitoring the
circuit and the AC line voltage. If a power failure occurs, the charger stops charging the battery, but the equipment does not notice the difference. The monitoring part of the system will start emitting beeps to let the administrator know
there is a loss of power.
Some UPS systems work with operating systems to perform shutdown
procedures for different types of servers.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

322

Chapter 8

Inside the Server Room

UPSs can be quite sophisticated. Most will attach to the server using a
serial cable and have some software that will allow for the shutdown. Some
will be sophisticated enough to use a modem in the server to send out a page
to a designated pager. Some UPSs come with plugins that will allow them to
work as SNMP devices or so they can be managed over the Web. Make sure
you look at what is important to your installation before buying a UPS, and
then tailor the UPS to your system.

Keyboard, Video, and Mouse (KVM)

We have all seen pictures in magazines and on TV of the computer room with
rows and rows of monitors piled high on top of one another, while a group of
people roll from side to side intently studying the screens to see what is going on
with the network. I have even been in a couple of data centers like that, and it
is really impressive, but that is not the normal state of affairs.
Without doing any scientific studies, I would hazard a guess that the
majority of data centers dont serve hundreds of thousands of people; they
will usually serve hundreds, or maybe a few thousand. While they all will
have some form of data center, usually it is not prime real estate, and it is
cramped and packed with equipment. So the thought of having banks and
banks of monitors is not only unreasonable, it is impractical.
The problem does remain that there are multiple servers. Each of those
servers has to be monitored somehow, and to monitor them requires a monitor, a keyboard, and in most cases, a mouse. This is where a keyboard,
video, mouse (KVM) switch comes into play.
A KVM switch will allow you to hook a multitude of computers to a single
keyboard, monitor, and mouse. Take a look at Figure 8.5.
FIGURE 8.5

KVM setup

PC1

PC2

PC3

PC4

Keemux-P4

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Cable Management

323

In this case, one KVM switch is controlling four servers. The administrator
can switch between computers by punching a button and switching from one
system to another.
These switches range from the simplistic A/B switch with two sets of video
cables, two sets of keyboard cables, and two sets of mouse cables coming in,
and one set going out, to KVM switches that can handle hundreds or even
thousands of servers from a single location. In the latter instance, the switch
itself is a network appliance, plugging directly into the network with a Cat 5
cable and communicating directly with each server.
If you have limited space, and you need to have more room, eliminate the
multitude of monitors with a KVM switch.

KVM switches can be somewhat problematic. They dont always work as

advertised. Sometimes turning them off and then back on again reminds
them who is boss and gets them communicating properly again.

Real World Scenario

Nobody ever said your life would be easy. Keeping track of cables is always
a challenge. First, there is the challenge of running the cables through hostile
environments with twists and turns and sharp objects and fluorescent lights
and other unfriendly things lurking unseen. Then, once you get the cable run
to the computer room, you have to figure out a way to get the cable to the
punch down block or to the connection device in some semblance of order so
the room at least appears to be neat and orderly. Then there is the challenge
of keeping track of where each individual cable comes from, just in case that
information is ever needed.
At the very least, keep cables off the floor and out of the way of cleaning
crews! This is especially true in work areas. I have seen many examples
where the cables lay in places where they could be run over by a chair, or
the cleaning crew would have to take special care not to get the cable
caught in the vacuum cleaner. This is one of those commonsense areas
that seems to get frequently overlooked.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

324

Chapter 8

Inside the Server Room

Updating Firmware

You are all familiar with hardware and software. If you have ever had
to update a Basic Input/Output System (BIOS), you have had to update
some Firmware.
If you are not familiar with the BIOS, it is the program that the microprocessor uses to figure out how to start up the computer. It is also responsible
for things like the way data flows between the operating system and all the
peripherals (such as video card, hard disk, keyboard, and mouse). The BIOS
is stored on an erasable programmable read-only memory (EPROM) chip,
and if something changes with the way the motherboard should be addressing a peripheral, the BIOS may need to get upgraded and the EPROM reprogrammed.
It is the Firmware that is the programming that is placed on the EPROM.
When you update an EPROM, like the BIOS, you are making a permanent
change to the computer. It is important to realize that the BIOS program is
not the only Firmware in your computer. There can be Firmware on your
modem, your printer, your video card, and even your disk controller. Each
of these may, at one point or another, need to be updated.

By the way, if you work in an IBM shop and you hear the term microcode, think
Firmware. IBM has its own way of discussing just about everything!

How Do You Know?

How do you know when to upgrade your Firmware? That is a really good
question and one that I wish I could answer with some degree of reliability.
After all, it is somewhat easy to tell when a Network Interface Card (NIC)
needs to be replaced, or when a hard drive has seen better days, but the question of Firmware is a little more difficult.
My suggestion on timing would revolve around when the manufacturer
issues new releases. In the next section, we are going to talk about (gasp!)
documentation, but it wouldnt be letting the cat out of the bag to say that
you need to know what makes up the major components of your server, and
every once in a while go out and check the Web site to see if anything has
changed. If it has, you have a decision to makedo you apply the revision
or do you wait.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Updating Firmware

325

This is not an easy decision. Some people believe firmly in the old adage,
If it aint broke, dont fix it. Others tend to be much more proactive, thinking that chances are manufacturers are not sitting around waiting for the
opportunity to tell their customers that something they sold doesnt work. If
the manufacturer does admit to a problem, it must be serious, and if they are
giving out a free fix to the problem, you might as well take advantage of it.
I have been over this argument dozens of times with proponents from
both sides of the question. As a matter of fact, I can do both sides, playing
devils advocate for or against. I call this flexibility. My wife and children call
it being argumentative. Whateverboth sides are very valid.
Because Firmware issues tend to be subtle, checking out the readme
before applying the upgrade would be beneficial. At that point, you can
decide if your system is suffering from any type of symptoms, and if it is, you
can apply the update and be done with it. If not, you may want to keep the
update handy, just in case something does decide to go south and you need
it in a hurry.

Applying Firmware Updates

Now, the nice thing about a software patch is that the patch will usually do
most of the work for you. If you are applying a service pack to a Microsoft
product, you just download the patch and click Setup.exe at the very most.
At the least, you find the service pack, decide to download it, and the installation wizard takes over from there.
Firmware patches are usually not that easy. Depending on what you are
updating, you may have to apply the upgrade from a diskette, a CD-ROM,
from a tape drive, or even from a serial port, using terminal emulation mode.
Since you are dealing with a very low-level part of the computer, this is something that should usually be undertaken with great caution. It is much easier
to uninstall a software patch than it is to uninstall a Firmware upgrade.
The designers of the Firmware upgrade, like the designers of a patch or
service pack, have the administrator at their mercy. They can make it really
easy on you, or they can make the upgrade process somewhat more challenging. For example, I have had to upgrade the BIOS on my server several times
in the last couple of years. The BIOS upgrade usually consisted of going out
to a Web site and downloading the upgrade, double-clicking the file name,
providing a floppy diskette, and then rebooting the computer with the Flash
BIOS upgrade diskette in the appropriate diskette drive. Not too bad at all.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

326

Chapter 8

Inside the Server Room

Compare that with the set of instructions in the sidebar, which came with
information for upgrading the Firmware on a Digital Equipment Corporation (DEC) Alpha workstation from a CD. I mean, for an old DOS prompt
kind of guy like I am, it does my heart good to see there are still times where
you have to enter commands!

Perform the following steps to update the Firmware from

the SRM console:
1. Insert the Firmware Update CD-ROM into the CD-ROM drive.
2. Type the show device command to determine the device ID of the
CD-ROM drive.
>>>show device
dka0.0.0.6.0 DKA0 RZ26L 440C
dka400.4.0.6.0 DKA400 TOSHIBA CD-ROM 0064
dva0.0.0.0.1 DVA0
ena0.0.0.1.1 ENA0 08-00-2B-38-67-6B
pka0.7.0.6.0 PKA0 SCSI Bus ID 7

3. Boot the Firmware Update Utility CD-ROM using the device ID display in
Step 2.
>>>b -fl 0,a0 dka400
Bootfile: [x]y.exe
(where x is the appropriate directory and y is the appropriate filename.) You
should see a display similar to the following while the update utility is loading and executing:
bootstrap code read in
base = 11e000, image_start = 0, image_bytes = ea800
initializing HWRPB at 2000
initializing page table at 110000
initializing machine state
setting affinity to the primary CPU
jumping to bootstrap code
ff.fe.fd.fc.fb.fa.f9.f8.f7.f6.f5.
ef.df.ee.ed.ec.f4.eb.....ea.e9.e8.e7.e5
X3.7-10895, built on Sep 13 1994 at 11:29:42

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Updating Firmware

>>>Execute Update Sequence

Update Script Complete
(boot pmem:180000 -flags 0,0)
bootstrap code read in
base = 180000, image_start = 0, image_bytes = 800000
initializing HWRPB at 2000
initializing page table at 710000
initializing machine state
setting affinity to the primary CPU
jumping to bootstrap code
apu-version 959328854 *** Firmware Update Utility V2.9 ***
*** System Type: Personal Workstation 433 ***
Update
Verify
List
Show
Dump
Verbose
Nverbose
Debug
Ndebug
?
Apu->
To update the Firmware, type update and SRM. The following information
will be displayed as the Firmware is being updated. The actual information
may change between versions.
Apu-> update SRM
APU-I *** ROM 0,1 Contents will be destroyed ***
APU-I Are you ready to Program Device? (y/n)y
APU-I Programming device
APU-I Programming device completed
APU-I Programming device
APU-I Programming device completed
Apu->

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

327

328

Chapter 8

Inside the Server Room

4. Use the verify command to verify the integrity of the SRM Firmware.
First enable verbose mode by typing verbose; then type verify and SRM.
Apu-> verbose
Apu-> verify SRM

5. To exit the update utility, cycle the power. Leave the power off for at least
45 seconds.

The bottom line of Firmware upgrades is to make sure that you have read
the readme for any installation tips before you start the upgrade!

Performing the Upgrade

Applying a Firmware upgrade will mean that your server is going to have to
be out of service, at least for a short period of time. Before scheduling the
outage, do the following:


Download the upgrade.

Read the readme to find out what is involved with the upgrade.

Make sure that you have all the software and hardware components
necessary to perform the upgrade.

Make sure that you know how to either reverse the procedure or go
back to your starting point. This may mean downloading not only the
latest upgrade, but also the previously applied upgrade, just in case.

When you schedule the outage, pay attention to the following:



Depending on the number of Firmware upgrades that you have to apply,

this should not be a very long process. EPROMs on servers do not hold
massive amounts of code, so the entire EPROM can be rewritten in a relatively short period of time. In this case, short period of time is measured usually in minutes. If you are resetting a BIOS, for example, it will
probably take longer to down the server and restart it than it will to
upgrade a BIOS. However, read the readme; it will give you an idea of
how long the upgrade can take. This is very important for Firmware
upgrades. There usually are not any really fancy screens to go along with
the upgrade to show the percentage completed. You may get impatient
and think the process should be over when it is not. Stopping a Firmware
upgrade in midstream is not a good idea!

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Updating Firmware

329

Even though you know it will take only a few minutes, leave yourself
plenty of time. Remember Murphys Law: Anything that can go wrong,
will go wrong at the worst possible moment. Remember also the corollary to Murphys Law that states, Murphy was an optimist.

After the upgrade, remember this advice:



Just because the upgrade took only a few minutes, dont rush off when
you are done. Stick around and make sure the server is operating
within specifications and nothing unexpected is occurring. This may
be a good time to break out your baseline of operations for that server
from the documentation.

Real World Scenario

Here is another tip, not only for taking the exam, but for running your network. Take a few minutes on a regular basis to visit the support section of the
Web sites of your hardware and software vendors. You would be amazed at
the things you will learn, and of the patches or fixes you will come up with.
There is another reason for frequent visits. Familiarity with the Web site will
help you know the strengths and weaknesses of each company. Some of the
major players in the hardware and software industry seem to have some serious trouble designing a Web site where people can actually find things.
Recently I bought a digital camera. When I bought the camera, the salesperson
assured me that USB drivers were available for my laptop operating system.
The salesperson lied. Several months later, I went back to see if the drivers
were available. I could locate documentation that said the drivers were available, but it took me about an hour to find the page where I could actually download the driver. When I did locate the page, I was using a specialized search
engine that I use for research, and the search was done away from the Web
site. The Web sites search engine would return hits for PCs, not digital cameras, and it never could figure out what a USB driver was.
When you have a problem with a system, check the vendors support Web
page. That should be the first place to look. If there are patches or BIOS
upgrades or just plain old fixes, try applying them before calling technical
support. As a former tech support rep, I can tell you the first thing the tech
support department is going to do is make sure that your system is up-todate on drivers, patches, and fixes before going too much farther.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

330

Chapter 8

Inside the Server Room

Documentation

When I was first starting out in this business, I did a stint as a telephone tech support representative for a major computer manufacturer. At
least they were a major computer manufacturer at the timethey have since
gone out of business. If you have never held that job, you probably dont
realize that it is a tremendous learning experience in troubleshooting and
communications. Whenever I would begin to troubleshoot an issue, the first
thing I needed to know was, What changed?

Govanuss First Law of Network Computing

End users lie. Always.

The answer I would invariably receive would be, Nothing. Then I

would have to plead, wheedle, cajole, beg, trick, scam, or otherwise pry out
of the user the admission that something had changed. This is where I
learned my greatest lessons about documentation. The more documentation
and change management you have, the easier it is to troubleshoot problems.
Every time I mention documentation, administrators in the room start to
giggle and titter. I can understand this. Documentation is one of those things
that are on the top of everybodys list of things to do. Unfortunately, it never
gets crossed off. It always stays there.
Documentation embodies all the things that we want to do, but we just
never seem to have time to do it! I am here to tell you that it is time to make
the time to do it! You would be surprised to know the ways that actually
having documentation can help you!

What to Document
What kinds of things should be documented in a well documented network
environment? The question should probably be, what kinds of things
shouldnt be documented in a well documented network? It would be a lot
easier to compile that list. Lets take a look at some of the things you may
want to have on record.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Documentation

331

Network Map
Call it a map of the network, a wiring diagram, whatever you want to call it,
you should have an overall map of the network. Depending on the size and
complexity of your network, this document may be several layers deep. For
example, look at Figure 8.6.
In this map we have a very-high-level overview that shows the cities that
our network connects. We can see there are T1 connections in the United
States, Asynchronous Transfer Mode (ATM) connections outside of the
United States, and an X.25 cloud connecting the rest. The X.25 cloud is handled by a WAN provider.
FIGURE 8.6

Top-level map of the network

Minneapolis

Hong Kong

X.25 Cloud
Orlando

Melbourne

Legend

Cloud
New York

Beijing

In U.S.: T-1
Outside U.S.: ATM
X.25 Services
provided by
XYZ Supplier

In the next level of the map, you would get more specific. For example,
look at Figure 8.7.
In this case we are showing the diagram of how each city connects to the
wide area network. We have specified the make and model of the router,
along with any pertinent IP information.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

332

Chapter 8

Inside the Server Room

The next level of map may be location specific. So inside the city of Minneapolis we have three locations. For each location I would list how they
interconnect, the number of users they serve, appropriate subnet information, appropriate router information, etc. My final level of the map may be
subnet specific. This is where you can get really granular.
FIGURE 8.7

Wide area network connections

Minneapolis

CDE Router
111.222.44.33
255.255.255.240

ZYX Router
222.111.33.43 Public
255.255.255.0
10.1.0.2 Private
255.255.255.0

Cloud

XYZ Router
222.111.33.44 Public
10.1.0.1 Private
255.255.255.0

Orlando

EFG Router
123.234.44.33
255.255.255.240

Lets assume that your network is laid out over three floors in a building.
Each floor has a wiring closet/server room providing services. Each floor has
at least a file server that can route traffic associated with it, and each floor
is its own subnet. In this case, you may have a blueprint of the floor, with all
the cubes laid out. You can spec out the network connections wired in each
cube, the phone connections in each cube, and how they are labeled in the
wiring closet. You can show where computers are located, where servers are
located, where printers are located, and where services like Dynamic Host
Configuration Protocol (DHCP) and Domain Name Service (DNS) are provided. For the DHCP server, you can specify which range of addresses this

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Documentation

333

DHCP server will provide, the DNS and Gateway information it provides,
the address of any Windows Internet Name Service (WINS) servers it provides, and anything else of interest. This document would be the first place
that someone could turn to if one of those servers went down. If the DHCP
server went down, if I could find the map, I could find the IP address that I
would have to assign to the new DHCP server as well as how it was configured. One-stop shopping, without hearing, I dont know! from a semiconfused IT person.

Server Configuration Documentation

For the server particularly and all workstations ideally, we want as much
configuration as possible. This means hardware, software, patch levels,
applications or services that are running, and anything and everything you
can get about the server, so if there is ever a question, you have the answer
at your finger tips. Depending on the type of hardware and operating system
software that you have available, this can be a relatively painless process.

Be sure to explore the Web site of the server operating system you are using
for tools that will make gathering this information easier. Any time you can
get the system to do the grunt work for you, the better.

I have several servers in my office. Because I routinely do work with both NetWare and Windows 2000 (and sometimes with Linux), I have a NetWare 5.1
server in my office that I use for a mail server and a couple of Windows 2000
Servers that I use for firewalls and other stuff. Novell has a nifty little
Config.NLM that can be loaded on just about any NetWare server and it gives
you your server configuration information stored in a text file format. Time to
complete the task, from download to printing, about 5 minutes. It was a 100K
file, so even with my slow ISDN link, it still took no time at all. While this is
decidedly NetWare specific, these are the kinds of things you really want to see
in your documentation. This file is 31 pages long in its native format, and while
I am sure you paid big bucks for this book so you could look at the config.sys
file of my server, I will spare you the agony. Here at least is the abbreviated version with just some of the stuff that was gathered.
Novell File Server Configuration Report For Server: PSC_MAIL
Novell File Server Configuration Report Created: Sun, Oct
15, 2000 8:18 am
Novell File Server Configuration Report. [Produced by
CONFIG.NLM v2.02]

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

334

Chapter 8

Inside the Server Room

Novell NetWare 5.00i May 15, 2000

(C) Copyright 1983-2000 Novell Inc. All Rights Reserved.
Server name...............: PSC_MAIL
OS Version................: v5.00
OS revision number........: 8
Server Up Time(D:H:M:Sec).: 4:13:43:20
Serial number.............:
Licensed Connections......:
Internal Net. Addr........: 05B44922h
Security Restriction Level: 1
SFT Level.................: 2
Engine Type...............: NATIVE
TTS Level.................: 1
Total Server memory.......: 207.63 MB or 217710592 Bytes
Processor speed rating....: 28889
Original cache buffers....: 52662
Current Cache Buffers.....: 22386
LRU Sitting Time(D:H:M:S).: 0:07:57:08
Current FSP's.............: 20
Current Receive Buffers...: 2000
Directory cache buffers...: 510
Workstations Connected....: 7
Max Workstations Connected: 9
Server language...........: ENGLISH (4)
**********************************************************
***********
Top of Modules List
148 Modules Loaded.
-------------------------------------------------------------------ADMSERV.NLM v5.10
May. 12, 2000 NetWare Web Manager
Copyright (c) 1997,1998, Novonyx, Inc., All Rights Reserved.
Copyright (c) 1999-2000, Novell, Inc., All Rights Reserved.
.
ZIP.NLM
v1.01g Jul. 20, 1999 Java Zip (based on 1.1.7)
(C) Copyright 1998 Novell, Inc. All Rights Reserved.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Documentation

335

-------------------------------------------------------------------End of Modules List

148 Modules Loaded.
**********************************************************
***********
**********************************************************
***********
Top of LAN Driver Configuration Listing
-------------------------------------------------------------------Signature.....:
CFG Version...:
Node Address..:
Board Number..:
Board Instance:
Media Type....:
MLID Version..:
Slot..........:
I/O...........:
Memory........:
IRQ...........:
DMA...........:
Logical Name..:

HardwareDriverMLID
1.14
00A0CC3613BF
1
1
ETHERNET_802.2
1.19
3
6800h -> 689Fh
None
10
None
LNE100TX_1_E82

Signature.....:
CFG Version...:
Node Address..:
Board Number..:
Board Instance:
Media Type....:
MLID Version..:
Slot..........:
I/O...........:
Memory........:
IRQ...........:
DMA...........:

HardwareDriverMLID
1.14
00A0CC3613BF
2
1
ETHERNET_II
1.19
3
6800h -> 689Fh
None
10
None

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

336

Chapter 8

Inside the Server Room

Logical Name..: LNE100TX_1_EII

-------------------------------------------------------------------End of LAN Driver Configuration Listing
**********************************************************
***********
**********************************************************
***********
Top of Storage Device Configuration Information
-------------------------------------------------------------------Storage Device Summary:
0x0000 [V025-A0] Novell ATA/IDE Host Adapter Module
0x0002 [V025-A0-D1:1] MATSHITA CD-ROM CR-581 1.05
0x0001 [V025-A0-D1:0] Maxtor 91000D8
0x0003
DOS Partitioned Media
0x0006
NetWare Partition
0x0009
Hotfixed Partition
0x000A
Non-Mirrored Partition [Group ID: 0x294A8AB1]
0x0004
Big DOS; OS/2; Win95 Partition
0x0005
NetWare Partition
0x0007
Hotfixed Partition
0x0008
Non-Mirrored Partition [Group ID: 0x294A8AB5]
Storage Device Details:
[V025-A0] Novell ATA/IDE Host Adapter Module
Media Manager object ID: 0x0000
Driver name............: IDEATA.HAM
.
.
.
DOS Partitioned Media
Media Manager object ID: 0x0003
Media type.............: IBM partition

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Documentation

NetWare Partition
Media Manager object ID......:
Activated....................:
Registered...................:
Functional...................:
Reserved.....................:
Logical partition............:
Beginning sector of partition:
Size, in sectors.............:

337

0x0006
Yes
Yes
Yes
No
No
5976180
13558860

Hotfixed Partition
Media Manager object ID...........:
Activated.........................:
Registered........................:
Functional........................:
Reserved..........................:
Logical partition.................:
Hot Fix offset....................:
Hot Fix identifier................:
Total Hot Fix blocks available....:
Used Hot Fix blocks...............:
Number of available Hot Fix blocks:
System Hot Fix blocks.............:

0x0009
Yes
Yes
Yes
No
No
67832
0x274CAB97
8479
1
8422
56

.
.
.
------------------------------------------------------------------End of Storage Device Configuration Information
**********************************************************
***********
***********************************************
*
Volume Statistics for SYS
*
***********************************************
Volume Size................: 1909 MBytes
Block Size.................: 64 KB
Total Blocks...............: 30558

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

338

Chapter 8

Inside the Server Room

Free Blocks................:
Purgable Blocks............:
Not Yet Purgable Blocks....:
Total Directory Entries....:
Available Directory Entries:
Sectors per Block..........:
Free Disk Space............:
Purgable Disk Space........:
Suballocation..............:
Compression................:
Migration..................:

11299
4125
1
152576
66087
128
706 MBytes
257 MBytes
ON
ON
OFF

***********************************************
*
Volume Statistics for DATA
*
***********************************************
Volume Size................: 6587 MBytes
Block Size.................: 64 KB
Total Blocks...............: 105397
Free Blocks................: 82698
Purgable Blocks............: 2815
Not Yet Purgable Blocks....: 0
Total Directory Entries....: 277504
Available Directory Entries: 238139
Sectors per Block..........: 128
Free Disk Space............: 5168 MBytes
Purgable Disk Space........: 175 MBytes
Suballocation..............: ON
Compression................: ON
Migration..................: OFF
***************************************************
Volume Name
Name Spaces Loaded
----------------- -------------------------------SYS
DOS
SYS
LONG_NAMES
DATA
DOS
DATA
LONG_NAMES

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Documentation

339

**********************************************************
***********
Top of file. [SYS:SYSTEM\ABEND.LOG]
-------------------------------------------------------------------File size is 0 bytes.
-------------------------------------------------------------------End of file. [SYS:SYSTEM\ABEND.LOG]
**********************************************************
***********
**********************************************************
***********
Top of file. [SYS:SYSTEM\TIMESYNC.CFG]
-------------------------------------------------------------------# TimeSync.Cfg is now updated automatically,
# when changes are made on the System Console
# TIMESYNC Configuration Parameters
Configured Sources =ON
Directory Tree Mode =ON
Hardware Clock =ON
Polling Count =3
Polling Interval =600
Service Advertising =ON
Synchronization Radius =2000
Type =SINGLE
# TIMESYNC Configured time source list
TIME SOURCE = 129.7.1.66:123
-------------------------------------------------------------------End of file. [SYS:SYSTEM\TIMESYNC.CFG]
**********************************************************
***********

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

340

Chapter 8

Inside the Server Room

**********************************************************
***********
Top of file. [SYS:ETC\NETINFO.CFG]
-------------------------------------------------------------------.
.
.
-------------------------------------------------------------------End of file. [SYS:ETC\NETINFO.CFG]
**********************************************************
***********
**********************************************************
***********
Top of file. [SYS:ETC\JAVA.CFG]
-------------------------------------------------------------------JAVA_HOME=SYS:\JAVA
OSA_HOME=SYS:\JAVA
MGMT_HOME=SYS:\PUBLIC\MGMT
CLASSPATH=sys:\java\lib\classes.zip
.
.
.
-------------------------------------------------------------------End of file. [SYS:ETC\JAVA.CFG]
**********************************************************
***********
**********************************************************
***********
Top of file. [SYS:ETC\HOSTS]
-------------------------------------------------------------------#
# SYS:ETC\HOSTS
#

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Documentation

341

# Mappings of host names and host aliases to IP address.

#
127.0.0.1 loopback lb localhost # normal loopback address
#
# examples from Novell network
#
130.57.4.2 ta tahiti ta.novell.com loghost
130.57.6.40 osd-frog frog
130.57.6.144 sj-in5 in5
192.67.172.71 sj-in1 in1
#
# interesting addresses on the Internet
#
192.67.67.20 sri-nic.arpa nic.ddn.mil nic
26.2.0.74 wsmr-simtel20.army.mil simtel20
-------------------------------------------------------------------End of file. [SYS:ETC\HOSTS]
**********************************************************
***********
**********************************************************
***********
Top of file. [SYS:ETC\RESOLV.CFG]
-------------------------------------------------------------------domain psconsulting.com
nameserver 2xx.xxx.xxx.xxx
nameserver 2xx.xxx.xxx.xxx
**************************************************
**********************************************************
***********
Top of file. [SYS:SYSTEM\AUTOEXEC.NCF]
--------------------------------------------------------------------

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

342

Chapter 8

Inside the Server Room

#The Support Pack install has moved timesync configuration

to sys:\system\timesync.cfg
SET TIME ZONE = CST6CDT
SET DAYLIGHT SAVINGS TIME OFFSET = 1
SET START OF DAYLIGHT SAVINGS TIME = (APRIL SUNDAY FIRST
2:00:00 AM)
SET END OF DAYLIGHT SAVINGS TIME = (OCTOBER SUNDAY LAST
2:00:00 AM)
;(Commented out by install...) SET DEFAULT TIME SERVER
TYPE = SINGLE
SET BINDERY CONTEXT = .OU=CORP.O=PSC.PSC_TREE.
.
.
.
-------------------------------------------------------------------End of file. [SYS:SYSTEM\AUTOEXEC.NCF]
**********************************************************
***********
**********************************************************
***********
Top of file. [SYS:SYSTEM\BSTART.NCF]
-------------------------------------------------------------------LOAD NWAIF103
.
.
.
-------------------------------------------------------------------End of file. [SYS:SYSTEM\BSTART.NCF]
**********************************************************
***********
**********************************************************
***********
Top of file. [SYS:SYSTEM\BSTOP.NCF]

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Documentation

343

-------------------------------------------------------------------UNLOAD NWINSEP
.
.
.
UNLOAD NWAIF103
-------------------------------------------------------------------End of file. [SYS:SYSTEM\BSTOP.NCF]
**********************************************************
***********
**********************************************************
***********
Top of file. [SYS:SYSTEM\C1ENV.NCF]
-------------------------------------------------------------------search add sys:\ni\bin
search add sys:\ni\update\bin
.
.
.
ENVSET JAVA_COMPILER=SYMCJIT
-------------------------------------------------------------------End of file. [SYS:SYSTEM\C1ENV.NCF]
**********************************************************
***********
**********************************************************
***********
Top of file. [SYS:SYSTEM\C1START.NCF]
-------------------------------------------------------------------load java
startx
SYS:\public\Mgmt\ConsoleOne\1.2\ConsoleOne.ncf

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

344

Chapter 8

Inside the Server Room

-------------------------------------------------------------------End of file. [SYS:SYSTEM\C1START.NCF]

**********************************************************
***********
**********************************************************
***********
Top of file. [SYS:SYSTEM\CONMAN.NCF]
-------------------------------------------------------------------java -classpath
\java\lib\classes.zip;\public\mgmt\rconj\slpsp.zip;\public\mgmt\rconj\rconsolej_resources.jar;\
public\mgmt\rconj\rconsolej.jar;\public\mgmt\Console1\lib\
manager.jar;\public\mgmt\Console1\lib\applets.jar;\public\
mgmt\Console1\ConsoleOne.jar;\public\mgmt\console1\Console
OneResources.jar;\public\mgmt\Console1\lib\appclasses.jar;
\java\lib\help.jar;\java\lib\swing.jar;\java\lib\jgl3.1.0.
jar com.novell.applet.Main
"sys:\public\mgmt\Console1\Shortcuts\Tools\Console
Manager"
-------------------------------------------------------------------End of file. [SYS:SYSTEM\CONMAN.NCF]
**********************************************************
***********
**********************************************************
***********
Top of file. [SYS:SYSTEM\DGNCMRL.NCF]
-------------------------------------------------------------------# dgncmrl.ncf [Version 1.00]
# Batch file to unload and reload the Digi Netware CAPI
driver
# DGNCMDRV.NLM. WHSMCAPI.LAN is unloaded so that
reinitialize
# system will reload WHSMCAPI.LAN for each interface and
cause
# it to regegister with DGNCMDRV.NLM

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Documentation

345

#
unload dgncmdrv
load dgncmdrv
unload whsmcapi
reinitialize system
-------------------------------------------------------------------End of file. [SYS:SYSTEM\DGNCMRL.NCF]
**********************************************************
***********
**********************************************************
***********
Top of file. [SYS:SYSTEM\DGNCMVER.NCF]
-------------------------------------------------------------------#
# NCF file to display NCM Driver component versions.
# Copyright (C) 1995-97, Digi International, Inc.
#
load verdump sys:system/dgncm*.nlm sys:system/adp.bin
-------------------------------------------------------------------End of file. [SYS:SYSTEM\DGNCMVER.NCF]
**********************************************************
***********
**********************************************************
***********
Top of file. [SYS:SYSTEM\ENVNISI.NCF]
-------------------------------------------------------------------###
### Java Environment setup for the Server Install
###
ENVSET
NICLASSPATH0=SYS:\NI\UPDATE\LIB;SYS:\NI\LIB;SYS:\NI\UPDATE
\DATA;SYS:\NI\DATA;
.
.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

346

Chapter 8

Inside the Server Room

.
ENVSET JAVA_COMPILER=SYMCJIT
-------------------------------------------------------------------End of file. [SYS:SYSTEM\ENVNISI.NCF]
**********************************************************
***********
**********************************************************
***********
Top of file. [SYS:SYSTEM\GATHERER.NCF]
-------------------------------------------------------------------LOAD ZENINV.NLM STR_DIR=DATA:\ZENWORKS\STRDIR
-------------------------------------------------------------------End of file. [SYS:SYSTEM\GATHERER.NCF]
**********************************************************
***********
**********************************************************
***********
Top of file. [SYS:SYSTEM\GRPWISE.NCF]
-------------------------------------------------------------------LOAD GWMTA @PSCDOM.MTA
LOAD GWPOA @CORP.POA
-------------------------------------------------------------------End of file. [SYS:SYSTEM\GRPWISE.NCF]
**********************************************************
***********
**********************************************************
***********
Top of file. [SYS:SYSTEM\GWIA.NCF]
-------------------------------------------------------------------search add SYS:\system

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Documentation

347

load SYS:\system\gwia @gwia.cfg-------------------------------------------------------------------End of file. [SYS:SYSTEM\GWIA.NCF]

**********************************************************
***********
**********************************************************
***********
Top of file. [SYS:SYSTEM\JEDITOR.NCF]
-------------------------------------------------------------------java -classpath
\java\lib\classes.zip;\public\mgmt\Console1\lib\novell_
jeditor.jar;\public\mgmt\Console1\lib\novell_common.jar;
\public\mgmt\Console1\lib\novell_classes.jar;
\public\mgmt\Console1\ConsoleOne.jar;\public\mgmt\console1
\ConsoleOneResources.jar;\public\mgmt\Console1\lib\appclas
ses.jar;\java\lib\help.jar;\java\lib\swing.jar\java\lib\jg
l3.1.0.jar;\public\mgmt\console1\resources\mach
\novell_jeditor_res.jar com.novell.applet.Main sys:
\public\mgmt\Console1\Shortcuts\JEditor
-------------------------------------------------------------------End of file. [SYS:SYSTEM\JEDITOR.NCF]
**********************************************************
***********
**********************************************************
***********
Top of file. [SYS:SYSTEM\JVORBCFG.NCF]
-------------------------------------------------------------------Envset CLASSPATH=$CLASSPATH;SYS:\JAVA\LIB\VBJORB.JAR
Envset CLASSPATH=$CLASSPATH;SYS:\JAVA\LIB\VBJAPP.JAR
Envset CLASSPATH=$CLASSPATH;SYS:\JAVA\LIB\VBJTOOLS.JAR
Envset VBROKER_ADM=SYS:\ETC
Envset OSAGENT_PORT=14000
--------------------------------------------------------------------

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

348

Chapter 8

Inside the Server Room

End of file. [SYS:SYSTEM\JVORBCFG.NCF]

**********************************************************
***********
**********************************************************
***********
Top of file. [SYS:SYSTEM\LDRCONAG.NCF]
-------------------------------------------------------------------LOAD RCONAG6 -E
-------------------------------------------------------------------End of file. [SYS:SYSTEM\LDRCONAG.NCF]
**********************************************************
***********
**********************************************************
***********
Top of file. [SYS:SYSTEM\MGRSTART.NCF]
-------------------------------------------------------------------bstart
load sys:odbc\odbc
load nwsqlmgr sys:\odbc\odbc.ini
-------------------------------------------------------------------End of file. [SYS:SYSTEM\MGRSTART.NCF]
**********************************************************
***********
**********************************************************
***********
Top of file. [SYS:SYSTEM\MGRSTOP.NCF]
-------------------------------------------------------------------unload nwsqlmgr
unload odbc
--------------------------------------------------------------------

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Documentation

349

End of file. [SYS:SYSTEM\MGRSTOP.NCF]

**********************************************************
***********
**********************************************************
***********
Top of file. [SYS:SYSTEM\NDS8.NCF]
-------------------------------------------------------------------nwconfig -b=sys:\\system\\nwupdate\\nds8b.ils e=sys:\\ni\\data\\err.log
-------------------------------------------------------------------End of file. [SYS:SYSTEM\NDS8.NCF]
**********************************************************
***********
**********************************************************
***********
Top of file. [SYS:SYSTEM\NISETSI.NCF]
-------------------------------------------------------------------search add sys:\ni\bin
search add sys:\ni\update\bin
search add sys:\java\bin
search add sys:\java\nwgfx
load java (CLIB_OPT)/B9
-------------------------------------------------------------------End of file. [SYS:SYSTEM\NISETSI.NCF]
**********************************************************
***********
.
.
.
**********************************************************
***********
Top of file. [C:STARTUP.NCF]
--------------------------------------------------------------------

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

350

Chapter 8

Inside the Server Room

LOAD IDEHD.CDM
LOAD IDECD.CDM
LOAD IDEATA.HAM PORT=1F0 INT=E
-------------------------------------------------------------------End of file. [C:STARTUP.NCF]
**********************************************************
***********
**********************************************************
***********
Top of file. [C:\CONFIG.SYS]
-------------------------------------------------------------------Files=30
Buffers=30
Lastdrive=Z
-------------------------------------------------------------------End of file. [C:\CONFIG.SYS]
**********************************************************
***********
**********************************************************
***********
Top of file. [C:\AUTOEXEC.BAT]
-------------------------------------------------------------------C:
CD \NWSERVER
SERVER
-------------------------------------------------------------------End of file. [C:\AUTOEXEC.BAT]
**********************************************************
***********

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Documentation

351

There is lots of information about the way the server is configured and the
way that all of the control files manage the server. Keep in mind that this file
has been severely pared down!
On the Windows 2000 side, there are also applications that you can use
like the System Information Files from the Computer Management
Microsoft Management Console. By going to Start  Program Files 
Administrative Tools  Computer Management, you can open the System
Tools and then open System Information. Once system information is highlighted, you can click Action  Save As System Information File and you will
have documentation of your server hardware and some of the software. If
you are not using Windows 2000, dont overlook the advantage of screenshot programs. Any time you write a book for Sybex, you are licensed to use
a program called FullShot 99, by inbit, inc. (www.inbit.com). With FullShot
you can do amazing things. Take a look at Figure 8.8.
FIGURE 8.8

Device Manager, Windows 2000 laptop

In this case, I just went into Device Manager on my laptop, opened the
area that might be important, and took a snapshot. I could then print that
out, and install it into my documentation.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

352

Chapter 8

Inside the Server Room

Your server documentation should include anything and everything that

you can think of that would be important if there were a problem. This can
include revision levels, services, patch levels, driver names, card manufacturers, slots the cards are in, IP addresses, and more.
Keep the information off site in a three-ring binder, on a diskette, or on a
CD, but keep the information! Once you have your original server documented, then it comes time to keep it up.

Maintenance Logs and Service Logs

A maintenance log is a listing of all the work you or your staff has done on
a particular system. These are usually routine, scheduled tasks. A service log
is a listing of all the service calls you have had to make on the system. In other
words, if I work for the company and I schedule an outage to upgrade the
Firmware, that is maintenance. Maintenance should be entered in the maintenance log and should include things like who was doing the maintenance,
what they were doing, why they were doing it, what changes were made,
what time the server came down, what time the server went back in service,
and what problems (if any) were encountered; there should be a spot for
observations. An observation would be something on the order of, When I
opened the server case, I noticed the top fan was very hot. We should keep
an eye on it to make sure it does not burn itself out.
If, however, I have to call my local hardware vendor to come out and
change out a hard drive because one has crashed, that is service. Any service
call should be entered in the Service Log and the entry should include things
like who was called, why they were called, when they were called, when they
arrived, how long they stayed, and whether the problem was fixed when they
left. A small distinction, but a distinction nonetheless.
How important are maintenance logs? Well, the first question you ask
someone whenever something has gone wrong is, What has changed?
I will be the first to admit, I am probably not the greatest person in the
world to be talking about documentation. After all, my documentation may
not be 100% up to date. But I can tell you how important it is.
A friend and I were working on a hardware problem. We had a NetWare
server that had been up for years suddenly start to abend. The server
had abended for three days in a row. The first day it was an oddity, so we
rebooted the server, it came back up, and life went on. The second day, we
paid closer attention to the abend message, and started talking about what
probably was going bad in the server to cause the abend. When the server
abended three days in a row, there was need for action! We started looking

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Documentation

353

at the message and decided that there was something wrong with memory,
so we were going to replace all the memory in the computer just to be safe.
As I walked back into the computer room to shut the lights off, I noticed that
there was a notepad lying next to the computer. The person who had discovered the original abend each of the three days had to be an administrative
assistant. She had made extensive notes on what the screen said so that,
when she went back to her desk to call us, she would have all the information
we might want. We had looked at the screens when we came in, but we had
never compared what each of the screens had said until we saw the notepad.
Suddenly, the fact that the server had abended three days in a row, at the
exact same time each day (well, within 15 seconds), changed our troubleshooting outlook immediately. It went from server hardware to environment. Sure enough, someone had installed a device in the room next door
that kicked off each day at the same time, and that device was on the wrong
circuit. When it kicked in, it pulled enough power to down our server.
If you make a change, add it to the maintenance logs. Then you can check
and see what was done recently. If there are problems, you will have the
information you need at your fingertips to make an informed decision about
the problems. You will also be able to rule out different troubleshooting
techniques because they have already been used.
Maintenance logs and service logs are important to the staff IT team, but
they are essential if you bring in a consultant. Often several people are dispatched over the course of a few days to troubleshoot a problem. Without
maintenance logs or service logs, each consultant goes back to square one
and spends time (and billable hours) relearning your system. It can be a very
expensive proposition.
The moral of the story is, if you make a change, log it! Keep the log where
it is instantly accessible. If you call in a consultant, make sure the consultant
logs her work.

When the bill comes for the consultant or service engineer, you may want to
go back and check it against the service log. As a former service engineer,
senior network hardware engineer, or whatever the heck my title was, I can
tell you that as a group, these people are not the most paperwork-proficient.
Sometimes they forget to do it, or lose it altogether. This is an inconvenience
to the engineer and the engineers boss when it comes time to submit utilization reports. However, if I have to re-create work orders, I may get the start
and the end time of the time on site wrong, and you, the customer will be hit
with a larger than expected bill. Having documentation can save you money.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

354

Chapter 8

Inside the Server Room

Real World Scenario

Documentation is a wonderful thing, and it comes in so many different formats.
One of the things that you should have available is what is referred to as a baseline study. A baseline study of your network basically lays out how the server
or network is operating under normal conditions. Now, what are normal conditions? Normal conditions are readings taken at various times of day, various
days of the week, various weeks of the month, and averaged, more or less. For
example, if you took all your readings Monday morning at 9:00 when everyone
is coming into work in the morning, you would see entirely different levels of
traffic than you would see later in the afternoon. By taking your readings at various times, it helps to average out the peaks and valleys and give you a more
realistic picture of your network.
Next, know what log files are routinely generated by your operating system, by
some devices, or by services that you have running on your server. Make yourself familiar with the log files, and what they contain. Some log files will contain
information that might be very interesting to the developers, but to you and I,
they make no sense. Other log files will be very straightforward and easy to
understand. If you know what log files are available, you will have an easier
time checking them if a problem arises.

Installing Hardware Using ESD Best Practices

Using good electrostatic device (ESD) practices has several practical

applications to it:


Use static bags whenever youre carrying a card, motherboard, RAM

chips, and other electrostatic devices from one place to the other
regardless of whether the room is antistatically protected.

Use an antistatic wrist strap whenever youre working inside computers.

Properly ground out equipment (which typically means that you have
grounded wiring that you can plug equipment into and that the room
youre running the computers in is properly grounded).

Power off and unplug devices when you work on them.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Installing Hardware Using ESD Best Practices

Use grounded rubber mats on your workbenches.

Monitor server room environment for excess humidity.

355

You can also wear antistatic vests when working on equipment. These
vests usually come with a coil cord that snaps into a snap on a rubber mat
at your workbench.
When working inside a server, remember to perform an orderly shutdown
of the system, unplug the computer, and wear, at a minimum, a wrist strap
when working inside the guts of the computer. Its preferable that you do not
work on a carpeted floor. If youre working with cards, drives, and other
equipment, make sure youre grounded and that the computer is grounded
before you remove the device from its static bag and install it in the computer.

Installing a UPS
Installing a UPS is fairly straightforward. Your first challenge is to find a
place for a heavy cumbersome piece of hardware close to the server youre
going to protect. Youll make sure that ample power is available for the
UPSthat you wont blow a circuit as soon as you power it up.
Ascertain that you have an available serial port thats open and eligible
for use by the UPS. Youll need it for the UPS software. Verify that the UPS
software can be used with your servers NOS.
Next you perform an orderly shutdown of the server and unplug it. You
plug all associated server power cords into the back of the UPS. Its typically
not wise to include extension cords or power strips in UPS devices, but Ive
seen it done with no problems.
Next you run a serial cable from the UPS to one of the servers serial ports.
If you dont have an open serial port, you wont be able to utilize the UPS
software and obtain alerts from the UPS.
Power up the UPS, then power up the server. When the server comes back
up, go in and install the UPS software and configure it.

Installing External Devices

Installing external devices can be painfully annoying and problem creating,
or can be as simple as you please. Some devices cannot be unplugged while
the server is runningor at best you have a hit or miss proposition. Depending on the manufacturer and version of NOS youre using, you may or may
not be able to replace the keyboard while the NOS is running. Windows NT
and 2000 Server allow this.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

356

Chapter 8

Inside the Server Room

Most other external devices will require you to perform an orderly shutdown on the system, add the device, then power back up and configure. Be
prepared to perform several restarts in the event that you run into an IRQ or
memory conflict. Ive had occasions where installing a piece of hardware
seemed really straightforward, only to spend several hours fighting a system
whereas other times a piece of hardware I thought would be a real challenge
went right in and was operational within moments. You just never know. Its
all about the "personality" of the server, how good the driver software is, if
youve got any memory, DMA, or IRQ conflicts, and so forth.
As a general rule of thumb, you should plan on performing device additions during routine maintenance windows. This is not something youll
typically do during working hours unless youre working on a new server
that no one is going to connect to.
Universal Serial Bus (USB) is a godsend but it wont work natively with
Windows NT Server without additional software. USB will work with Windows 2000 Server and provides you an easy way to add hardware via a USB
port and configure real-time during working hours without having to reboot
the server. Unfortunately, youre not going to be able to buy too many
server-class external devices that use USB.
Remember that, with external SCSI peripherals, you may or may not have
to terminate the device. It depends on the ID number its using and where it
falls in the chain. Remember that the first and last SCSI IDs are the ones that
must be terminated.

Configuring External Peripherals

o matter what external device youre installing, you generally have a

configuration step you have to go through. Generally the device youre
installing will come with a software CD that has the device drivers you need
to run the device through your NOS. Its usually a pretty good idea to go to
the Web and download the latest and greatest device drivers for the product
(or at least verify that your CD software is the latest and greatest) before
installing the device. Generally the manufacturers will release updates to
device driver software that fix minor glitches. In most cases, updated device
driver software has helped me solve problems and has never created any
problems for me.
You may also have to configure a device once its installed, visible to the
NOS, and properly working. For example, suppose that you install a DSL

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Upgrading a UPS

357

modem in a server. You may have to go in and configure some DSL settings
on the modem so it can properly communicate over your DSL line.
Youll most likely run into problems with external peripherals if youve
run out of IRQs and your device is trying to use an IRQ thats already in use,
if you have memory conflicts with something thats already using the memory address your device would like to use, or if youre using more SCSI IDs
than are supported for the version of SCSI youre using. All of these issues are
solvable, but may create long configuration times while you figure out the
issue and then correct it.

Upgrading a UPS

epending upon the size of the UPS, upgrading a UPS typically

amounts to replacing the batteries in the old UPS or simply performing a
forklift upgrade where you replace the UPS entirely. There just arent that
many parts to a UPS. In some cases you may be able to upgrade the firmware
for the UPS interface but beyond that youre limited.
With room-size UPS interfaces, there are many different features and
add-ons that you can choose to add or upgrade. Consulting with vendors
and determining the type of service features that you want will result in a
somewhat customized configuration with your UPS.
The problem with upgrading a UPS is that you have tons of computing
gear hooked to it that you need to gracefully shut down before upgrading.

Performing an Upgrade Checklist

When considering any kind of upgrade, you should prepare an upgrade
checklist. Items on this checklist will include things like reviewing the production documentation, readmes, Web information, and other information
relative to your upgrade. Your checklist should also include the scheduled
date and time for the upgrade and list of steps that youll need to go through
to perform your upgrade. Keep in mind that the servers hooked to the UPS
(along with any externally powered peripherals hooked to the UPS) will have
to go through an orderly shutdown before you can upgrade.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

358

Chapter 8

Inside the Server Room

Recognizing and Reporting on Physical

Security Issues

ts amazing to me how lax security can be when it comes to servers and

server rooms. Companies are basing their entire operation on servers and yet
the server rooms are unlocked and the server is logged in 24x7 with admin
privilegeswhats up with that?
As a server admin, it is always important that you consider the safety and
security of the servers within your charge. Of special importance are the
physical and environmental settings your servers are situated in.

Limiting Access to the Server Room

I once worked for a company that had an electronic box on their server room
door. You had to key in three numbers and swish, the electric door glided open
and you gained entry. It was a pretty slick system. With one major limitation:
Everyone in the company knew the pass code. For all I know there may have
been people on the street that knew the code as well. The code was known by
people not even remotely associated with servers or computers.
On top of that, the code was never changed. Its the same today and Ive
been away from this company for more than three years! Do you see any
problems with a scenario such as this?
Server rooms are not places where just any Tom, Dick, or Mary should be
allowed to go. Only server admins and key designates should be allowed anywhere near a server room. The room should be locked with either a card-key
system (where only certain cards have access to the room) or with standard
door keys.
Backup tapes should be stored in the same place as the servers because
they are then protected by air-, power-, and humidity-conditioning and
theyre away from prying eyes and hands.

Assuring Physical Locks Are on Doors

Heres an interesting thing: Microsoft obtained a class C2 security rating
with their Windows NT 4 software. But were you aware that part of the C2
description says that server equipment must be behind a locked door? In
other words, take the locked door away and NT 4 is no longer considered C2
secure! This is an example of how critical your thinking must be relative to
your servers.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Recognizing and Reporting on Server Room Environment Issues

359

Ensure that your servers are behind good solid locked doors and that only
server administrators and their managers have access to the room. For this
reason, oftentimes telephone equipment is kept in a different room from the
servers.

Establishing Anti-Theft Devices

If you keep your servers in racks, you can lock the doors to the racks, thus
keeping people from stealing cables, inadvertently booting a server, and
other untoward situations that youd rather not have happen.
Keep in mind that operators and others who might do routine maintenance in and around server racks (cable folks, for example)people who are
accustomed to working around high-cost server gearshould know where
the keys to the racks are. But that doesnt mean that the general public
should be able to open rack doors and mess around.

Recognizing and Reporting on Server Room

Environment Issues

server rooms environmental issues are also importantin some

ways perhaps even more so than the security issues. There are several categories we need to speak of when we talk about environmental concerns.
Temperature It is incredibly important to maintain a good even temperature in the high 60s to low 70s (Fahrenheit) in your server room. Servers put
off tremendous amounts of heat and you have to have a way to dissipate it
somewhere so that the servers dont overheat. When servers overheat, they
get really funny and eventually stop working entirely. It is not a pretty situation. You need to be extremely concerned about the server room temperature
and monitor it daily.
Humidity Equally important is the need to regulate a server rooms
humidity. Too humid and youll wind up with electrostatic problems that
you dont need. The room cant be too dry, but it can be too humid.
Installing a humidistat is a good idea in any server room and a dehumidifier may be in order for more humid climates.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

360

Chapter 8

Inside the Server Room

ESD "Formal" server rooms typically are built on an elevated floor

where the air-conditioning is pushed through the underfloor and moves
out through holes in the floor tiles. One of the reasons for this careful layout has to do with electrostatic interference on the computing and associated peripheral and infrastructure gear. The flooring framework is
usually grounded out with stout copper wire going to a heavy earth
ground, thus providing you with a very stable ESD platform to work
from. Workbenches in server rooms should still be protected with rubber
mats that are grounded. You should still use wrist straps when working
on server gearthough to be truthful, theres not an admin I know who
does use them.
Power surges You can and should also consider equipping your server
room with power surge equipmentwhether the solution consists of a
room-size unit that costs tens of thousands of dollars or individual surge
protectors for each computer. A single electrical spike to a computer can
immediately trash the motherboard and render the computer useless. Or,
even worse, a spike can do some minimal damage to a circuit board and
cause irregular behavior that you might never be able to pinpoint.
Back-up generators Youre getting pretty sophisticated when you begin
adding a backup generator to your server room so that in the event of a
local power company blackout, your servers continue to operate. Ive
worked for companies that had their entire server room hooked to a room
full of car batteries. The batteries would keep the room up for a time and
then, if the outage was still in effect, the backup generator would kick in.
It was a diesel-fired generator and it would stay operational as long as a
maintenance worker would keep it stoked with fuel.
How serious do you have to be? Well, that depends on how reliant your
company is on its servers and computing gear. For an e-commerce company
a blackout might be the kiss of death. For a manufacturing concern it may
be no big deal (because the machines are down too). This is a decision you
and your management have to make. The decision to go forward with
backup generator power could cost hundreds of thousands to implement.
Question: How much would your company lose in terms of dollars if it
were down for, say, 24 hours?
Fire suppression Fire suppression can present problems to you when
youre considering environmental concerns. You cant install a sprinkler system in a room full of electronics. Forget about the time when the system goes
off and floods all the computers with waterwhat about the time when it
springs a leak and takes out your e-mail server? No, water will not do.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Performing Physical Housekeeping

361

Halon, on the other hand, is bad news for humans. You cant be in a server
room when halon goes off or youll wake up dead. You only have a few seconds to find the door and exit when a halon fire-suppression system goes
off. So halon is no longer vogue for computer rooms.
Several companies that specialize in fire-suppression equipment are going
with an alternative to halon called FM200 (heptafluoropropanealso
known in the industry as HFC-27ea). You can find lots of fire-suppression
companies on the Web simply by doing a search on FM200.
Flood concerns Lastly we think of flood concerns. Server rooms generally
should not be located in basements or on the first floor of buildings that are
close to rivers, lakes, streams, or other bodies of water that could potentially
flood into your building. In most cases, professionals typically recommend
that you put server rooms on a second or third floor so theyre away from
flood danger. If youre in an earthquake zone and close to an area capable of
flooding, then putting a server room on the second floor may be a bit more
sketchy of a consideration. In either case, you have to ask yourself the
question about flooding and how to guard against it.

Performing Physical Housekeeping

eeping your server room neat and clean is highly important. For starters
theres the safety issue that you have to be concerned with. If youve got the
covers to servers and racksthings like thatlying around, youve got the
potential for knocking something over on yourself, or cutting yourself.
Cabling should be neatly bundled, labeled, and tucked away in chases,
innerduct, or other method of keeping cabling in a straight line and away
from feet.
Servers should be screwed into their racks. I had a friend who never
screwed a server fast to its rack. It was very easy for someone to come up and
pull a server out and potentially disrupt some of its cabling.
Workbenches should be kept neat and orderlytools should be put back
in a storage box when youre done with them. Nuts, washers, and other
attachment devices need to be kept in storage containers.
Keeping a nice tidy server room will provide you with much faster reaction
times during trouble, will keep people from pulling cords out of machines and
walls, and will make for a more professional-looking implementation.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

362

Chapter 8

Inside the Server Room

Summary

This chapter focussed on how to do things right: how to string wire,

how to minimize the impact of upgrades, and how to keep track of changes
to your network. All of these things will help you achieve the long-term goal
of 99.99999% uptime.
This chapter covered a variety of hardware-oriented issues. For example, you
should use EST best practices when installing internal or external devices such as
boards, drives, processors, additional memory, internal cabling, etc. Never
transport circuit devices without static bags. Use wrist straps. Use rubber mats
when working on computers. Unplug servers when working on them (after performing an orderly shutdown, of course).
We talked about installing and upgrading UPS equipment. UPS devices
have batteries that youll need to monitor. They can hook to serial ports on
servers and you can install UPS monitoring software that will allow you to
set up alerts for things such as power brownouts, UPS equipment getting hot,
and so forth.
We also talked about configuring external peripherals. Youll need to be
concerned about SCSI terminations with SCSI equipment, and also that you
dont exhaust the allowed number of SCSI IDs.
We mentioned security issues such as being sure to lock server room doors
and racks. We covered environmental issues such as monitoring heat, humidity, providing UPS and backup generator equipment and watching out for
flooding potential. We also talked about how important it is to keep up with
the housekeeping in your server room to prevent accidents, to prevent clumsy
souls from walking on your gear, and to provide a neat professional-looking
environment.
I have spent the first eight chapters talking about different types of servers and
different hardware. In Chapter 9, Network Operating Systems, we are going
to look at the different types of server operating systems. We will look at the systems that are at the top of the heap, Microsoft Windows NT/2000 and Unix, at
operating systems that are seeing their popularity decline in Novell Netware,
and at an operating system that did not gain widespread acceptance in the local
area network or wide area network environment, IBM OS/2.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Summary

363

Exam Essentials
Know what a KVM switch is and how to use it A KVM switch allows
you to hook multiple computers up to a single monitor, mouse, and
keyboard.
Know how to get cables up off the floor Keep cables away from normal,
everyday office hazards like chair wheels and cleaning crews.
Know the proper layout of rack mounted devices Place heavy stuff at
the bottom to give the rack an anchor and make sure it does not fall over.
Know how the size of rack mounted devices is determined Rack
mount devices are measured in Us. Each U is 4.445 centimeters, or 1.75
inches.
Know how to secure a rack Be sure the rack has a door and make sure
the door locks.
Know when BIOS or Firmware upgrades should be performed When
there are problems, check the vendors Web site to see if there are new
upgrades, patches or fixes. If there is no problem, check the vendors Web
site anyway, because upgrading the BIOS or the Firmware is a form of
proactive maintenance.
Know where to get BIOS or Firmware upgrades The vendors Web site
should be your friend! Visit often and get to know it well.
Know what types of information should be put into maintenance and
service logs If it has changed, has been added, or is acting up, it should
be logged.
Employ ESD Know and understand how to install hardware using ESD
best practices.
Know how to install a UPS Know the steps involved in installing or
upgrading a UPS.
Be able to configure external devices Know whats involved in configuring external devices such as keyboards, monitors, subsystems, and
other equipment.
Recognize physical security issues Recognize physical security issues
such as physical locks on doors and providing anti-theft equipment for
racks and other devices.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

364

Chapter 8

Inside the Server Room

Recognize environmental concerns Recognize environmental concerns

for server roomsthings such as temperature, humidity, ESD, power surge
protection, backup generation, fire suppression, and flood considerations.

Key Terms
Before you take the exam, be certain you are familiar with the following terms:
Asynchronous Transfer Mode (ATM)
backbone
Basic Input/Output System (BIOS)
cable management
cable tray
conduit
Electromagnetic Interference (EMI)
Electronics Industry Association (EIA)
erasable programmable read-only memory (EPROM)
Ethernet
fiber optic cables
Firmware
footprint
horizontal cable
keyboard, video, mouse (KVM)
maintenance log
microcode
Network Interface Card (NIC)
plenum
raceway
rack
service log
standby power supply (SBS)

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Summary

Telecommunications Industry Association (TIA)

terminal emulation mode
Token Ring
uninterruptible power supply (UPS)
wiring diagram

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

365

366

Chapter 8

Inside the Server Room

Review Questions
1. Essie has been charged with installing a Jaz drive in one of the servers

for the purpose of sharing out 100MB disks that have drawings on
them that are created by engineers and need to be viewed by quite a
few people across the net. The goal is to install the Jaz drive, then share
the files so that they can be viewed by a certain group of users. What
external connections does Essie have at her disposal?
A. SCSI
B. Serial
C. Parallel
D. PCI
E. USB
F. Firewire
2. The company you work for has a large Internet presence. The company

has a staff of Web developers. One day youre told by the developers
supervisor that they will need to get into the server room from time to
time to check on the status of a server, perhaps to reboot it or reconfigure ODBCat a very minimum they will need to have the screen
logged in at all times. What are some answers you might give this
supervisor?
A. Developers will have to be accompanied by an admin at all times.
B. Developers are free to come and go as necessaryheres the code

to the door.
C. Developers can e-mail any configuration changes to the admins

and theyll take care of it.

D. Developers can use a remote control tool to get into the server and

make changes.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

367

3. Ananth has a server thats attached to a two-and-a-half-year-old

UPS. The UPS hasnt give him a bit of trouble over the years, but hes
beginning to suspect that it is reaching the end of its product life
cycle and may need to be upgraded or replaced. What are some
things that Ananth could consider upgrading in his UPS to boost its
life a little more?
A. Batteries
B. Add a serial card
C. UPS software in the NOS
D. UPS BIOS
4. What are some server room environmental concerns that a diligent

server admin would pay attention to?

A. Room EMF
B. Room temperature
C. Room humidity
D. Room noise
E. Room power conditioning
5. What are some components of physical housekeeping in a server room?
A. Keeping cables and power cords out of the way and neatly bundled
B. Providing uninterruptible power and power conditioning
C. Ensuring that racks are even, bolted together, and that rack doors

are regularly closed (and locked if necessary)

D. Installing a dehumidifier for too-humid rooms

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

368

Chapter 8

Inside the Server Room

6. Nell is a server administrator for a large telecommunications corpo-

ration that has 75 NT and 25 UNIX servers in its server room. She has
been charged with the physical aspects of the server room. The UPS is
a big room-sized unit that can service the entire server room. The UPS
is at maximum load right now, but when Nell checks into the problem
she finds out from the vendor that the UPS can be upgraded. When
preparing her upgrade checklist, what are some items that she needs to
jot down so shes sure to remember the detail items?
A. What is the cost of the upgrade?
B. Are there BIOS upgrades?
C. Ask the question, "Should the UPS be replaced?"
D. What are the steps I should take as I go through the upgrade?
7. Which of the following can be hooked up to a KVM switch?
A. Server
B. Monitor
C. Keyboard
D. Mouse
E. Modem
8. Where is Firmware stored? (choose the best answer)
A. BIOS chips
B. EPROMs
C. ROM
D. RAM
E. RAD
9. For Firmware to be upgraded, the server should be temporarily

removed from service.

A. True
B. False

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

369

10. Which of the following would not be considered a Firmware upgrade?

A. Applying a service pack to a Windows NT installation
B. Upgrading the BIOS on a server motherboard
C. Upgrading the code on an EPROM on a video card
D. Reprogramming the BIOS on a SCSI controller
11. How long would you normally schedule for a single Firmware

upgrade?
A. An entire weekend
B. An entire day
C. Hours
D. Minutes
E. Weeks
12. What is a good source of information about the tasks the Firmware

upgrade will accomplish and how long it will take?

A. www.firmware.com
B. The documentation that came with the server
C. The documentation that came with card
D. The documentation that came with the rack
E. The readme file that came with the Firmware upgrade
13. When doing a Firmware upgrade, what else should you do?
A. Be prepared to roll back the upgrade to the current levels if there

are problems.
B. Clean out the dust bunnies from inside the server.
C. Oil the fan in the server.
D. Add more disk space to the server.
E. Add more memory to the server.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

370

Chapter 8

Inside the Server Room

14. What is the difference between a maintenance log and a service log?
A. Maintenance is usually done by an outside vendor, but service is

done internally.
B. Maintenance is hardware related, but service is software related.
C. Maintenance is usually done the local staff, but service may be

done by an outside vendor.

D. Maintenance is planned, but service may or may not be.
15. What kinds of things should be in your server documentation?
A. Date the server was put in service
B. Who the server was purchased from
C. Length of warranty of the server
D. The server configuration
E. Operating system
F. Patches applied to server operating system
G. All of the above
16. Who should be in charge of documentation?
A. IT manager
B. Documentation manager
C. Senior technician
D. Whoever makes the changes
17. Documentation is one of those tasks that makes being in the field of

Information Technology worthwhile.

A. True
B. False

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

371

18. What are two types of system logs?

A. Room logs
B. Maintenance logs
C. Operating system logs.
D. Service logs
19. Using fiber optics as a cable backbone is preferable over unshielded

twisted pair because of the following:

A. Lower signal attenuation and higher resistance to EMI
B. Higher signal to noise ratio
C. Increased susceptibility to EMI
D. Price
20. Identify two types of cable management system.
A. J-hook
B. Ladder
C. Hook and ladder
D. Plastic ties

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

372

Chapter 8

Inside the Server Room

Answers to Review Questions

1. A, B, C, E, F. PCI is a bus-type connection, but all the others are

valid. Of the above list the best choice is probably USB because its
fast, easy to install, and instantly recognized by systems that support
USB. Typically it doesnt require a reboot of the system, though you
may have to check the documentation for your system.
2. A, D. This is a real-life scenario that actually happened at one place

where I worked. The developers were given door codes and could come
and go as they pleased. Within six months time the servers were so
goofed up it took months to get them back to solid operations. You
should insist that an admin is with the developers at all times, watching
them like a hawk, or that theyre allowed to remote into certain boxes.
But you should never allow open access to non-admin types.
3. A, D. Depending on the interface on the UPS, Ananth may be able

to upgrade the systems BIOS. He may also be able to replace and

even upgrade its batteries. Adding a serial card is probably out of
the questionit seems that UPS devices dont have a lot of add-on
components you can add. Upgrading the UPS software in the NOS
doesnt really help the UPS.
4. B, C, E. The servers are likely throwing off some EMF, but its not

something you can control; neither is the rooms noise. Youll find lots
of servers in a room to be extremely noisy. However, ensuring that
servers are delivered precise voltage (power conditioning) and that the
rooms temperature and humidity levels remain static are important
environmental concerns.
5. A, C. Physical housekeeping simply means that you keep your server

room in order, neat, picked up, and uniform looking. You definitely
want to keep cables and power cords out of the way because, even if you
dont have big feet, the lummox who unplugs your e-mail server from
the wall may definitely have them! Rack placement, the bolting together
of racks, and keeping rack doors shut provide for a less cluttered, more
professional looking environment. Setting up a dehumidifier solves an
environmental concern, but isnt considered housekeeping.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Answers to Review Questions

373

6. B, D. When writing an upgrade checklist, Nell probably already

covered the question in option C and she has decided to upgrade.

Option A probably drove the decision to upgrade rather than
replaceso its not really a candidate for an upgrade checklist
either. But Nell definitely wants to know if there are firmware
upgrades she needs to perform on the systems BIOS (if there is
one). Shell want to write down the steps that she (or the vendor)
intends to take as the upgrade goes forward. The checklist should
also include reading up on the systems documentation, Web site
information on the device, readme files, and so forth.
7. A, B, C, D. You can hook multiple servers or computers to a KVM

switch so they can share monitors, keyboards, and mice.

8. B. Firmware is stored on EPROMs.
9. A. Upgrading Firmware will require the server to be powered off, Firm-

ware changed, and the server to be restarted. It is best to take the server
out of service temporarily to perform the upgrade.
10. A. Applying a service pack is not a Firmware upgrade.
11. D. The actual Firmware upgrade will usually take minutes.
12. E. Consulting readme files can save you hours of frustration spent trying

to figure things out yourself or waiting on hold with customer service.

13. A. Whenever you do an upgrade, you should be prepared to roll back

the upgrade in case it causes more problems than it fixes.

14. C, D. While this truly is just semantics, maintenance is usually planned,

but service is usually caused by an unexpected outage. Maintenance is

usually done by the internal IT department and, depending on the severity of the problem, service may be done by an outside vendor.
15. G. All of this and more should be placed in the server documentation.
16. D. Whoever makes changes to the system should be the one to enter

the changes in the logs.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

374

Chapter 8

Inside the Server Room

17. B. Documentation is usually not thought of as the most fun a person

can have, but it is a necessary part of any IT department.

18. B, D. The two types of logs covered in this objective are maintenance

logs and service logs.

19. A. Fiber optics can send a signal over longer distances, with less atten-

uation and higher resistance to EMI. Therefore it is a better choice for

longer backbone cable runs.
20. A, B. The two types of cable management system talked about in this

section were J-hook and ladder.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Chapter

Network Operating
Systems
SERVER+ EXAM OBJECTIVES COVERED IN
THIS CHAPTER:

1.1 Conduct pre-installation planning activities.


Plan the installation.

Verify the installation plan.

Verify hardware compatibility with operating system.

Verify power sources, space, UPS, and network availability.

Verify that all correct components and cables have been

delivered.

2.3 Install NOS.



Configure network and verify network connectivity.

Verify network connectivity.

2.5 Install NOS updates to design specifications.

2.6 Update manufacturer specific drivers.

2.7 Install service tools (SNMP, backup software, system
monitoring agents, event logs, etc.).

2.8 Perform Server baseline.

3.2 Add Processors.


On single processor upgrade, verify compatibility.

Verify N 1 stepping.

Verify speed and cache matching.

Perform BIOS upgrade.

Perform OS upgrade to support multiprocessors.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Perform upgrade checklist, including: locate/obtain latest

test drivers, OS updates, software, etc.; review FAQs,
instruction, facts and issues; test and pilot; schedule
downtime; implement ESD best practices; confirm that
upgrade has been recognized; review and baseline;
document upgrade.

3.4 Increase memory.



Verify hardware and OS support for capability increase.

Verify memory is on hardware/vendor compatibility list.

Verify memory compatibility (e.g., speed, brand, capacity,

EDO, ECC/non-ECC, SDRAM/RDRAM).

Perform upgrade checklist including: locate and obtain latest

test drivers, OS updates, software, etc.; review FAQs,
instructions, facts and issues; test and pilot; schedule
downtime; implement using ESD best practices; confirm that
the upgrade has been recognized; review and baseline;
document the upgrade.

Verify that the server and OS recognize the added memory.

Perform server optimization to make use of additional RAM.

3.8 Upgrade system monitoring agents.



Perform upgrade checklist including: locate and obtain latest

test drivers, OS updates, software, etc.; review FAQs,
instructions, facts and issues; test and pilot; schedule
downtime; implement using ESD best practices; confirm that
the upgrade has been recognized; review and baseline;
document the upgrade.

3.9 Upgrade service tools (e.g., diagnostic tools, EISA

configuration, diagnostic partition, SSU, etc.).


Perform upgrade checklist including: locate and obtain latest

test drivers, OS updates, software, etc.; review FAQs,
instructions, facts and issues; test and pilot; schedule
downtime; implement using ESD best practices; confirm that
the upgrade has been recognized; review and baseline;
document the upgrade.

4.2 Create baseline and compare performance.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

ow, this is going to be a tough chapter for your favorite

geek author to get through. First of all, when the talk turns to major features, one persons feature is another persons curse. Secondly, this topic
borders on religion. While the debate between Novell and Microsoft has
pretty much died down, the debate between UNIX/Linux and Microsoft
Windows NT/Windows 2000 is still going strong. So, this chapter ought to
be good for classroom discussions if nothing else!
This chapter is going to be much more of a marketing type chapter than
many of the others. I am not going to compare and contrast the operating
systemsI am going to present their features. There used to be a time when
the goal was a single Network Operating System (NOS) environment, but
that has seemed to fade into oblivion. Most networks make use of several of
the operating systems listed above. Since each NOS has its strengths, it only
makes sense to take advantage of them when you can.
In addition, we are going to take a look at the load placed on a server
when you run directory services or a domain environment. That discussion
will be blended into the discussions of the various operating systems. I am
doing that just to make sure you are paying attention!
With that said, lets take a look at Novells NetWare.

For complete coverage of objective 1.1, please also see Chapter 6. For complete coverage of objective 2.3, please also see Chapter 10. For complete coverage of objective 3.2, please also see Chapters 3, 6, and 8. For complete
coverage of objective 3.4, please also see Chapter 4. For complete coverage of
objective 3.9, please also see Chapter 5.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

378

Chapter 9

Network Operating Systems

Novell NetWare

Novell NetWare is the venerable file and print server that has been
around for as long as there has been a local area network (LAN). In recent
years, Novell has been going through a redesign phase, attempting to discover
what its place is in the new era of networking.
This change has taken place across several fronts. Starting with NetWare 4,
Novell introduced NetWare Directory Services (NDS) to help network administrators manage enterprise networks. As NDS continued to mature, it was given
a new moniker for the e-business age; it is now called the NDS eDirectory.
Novell is trying to position itself in the business-to-business e-commerce
market with NetWare 5.1. It has bundled NetWare 5.1 with IBMs WebSphere Application Server 3 for NetWare. WebSphere provides extensive
Java support, which is now deeply integrated into NetWare.
NDS eDirectory is truly the lynchpin of the new Novell. It is what makes
NetWare suitable for use in a large organization. NDS has been around for
years, so as a directory service, it is proven and has a track record. More
importantly, as a mature product, it has been broken and has been fixed, and
has a well trained and experienced workforce to run the network.
Administration of a NetWare network is getting easier. ConsoleOne is a
Java-based administration tool that can be used to remotely administer your
network. It can also be used with the NetWare Management Portal.
Lets take a look at some of the features of NetWare.

WebSphere Application Server

As I mentioned above, the IBM WebSphere Application Server 3 for NetWare is integrated into NetWare 5.1. The Application Server is protected
with Secure Sockets Layer (SSL) 3 security. This way, you can use the Web
server to develop enterprise-wide Web-based applications and do it securely.
The WebSphere Application Server is Java-based and it includes a set of
development tools called WebSphere Studio.
WebSphere also comes with management tools that will let you analyze
performance and usage information about the Web applications that you
have developed. WebSphere Application Server has a runtime servlet engine
that can handle requests for database information. Because the servlets are
written in Java, they are protocol and platform independent, so they will
work with any existing hardware and software.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Novell NetWare

379

The Application Server has full client/server capabilities, so that an information request that is made up of a series of complex requests can be offloaded
from the servlet directly to the Application Server. Once the Application Server
has processed the request, it will return the information through the Web
server to the Web browser. Since the Web server is taken out of the loop and
does not have to search through other applications databases, it can handle
the more mundane queries quicker. This will improve the overall performance
of the network.

NetWare Management Portal

The NetWare Management Portal allows system administrators to manage
their IP-based NetWare 5.1 servers from any Web connection on the network.
The Management Portal is installed and loaded during the server installation
process. The front door to the portal is shown in Figure 9.1.
FIGURE 9.1

NetWare Management Portal

This means that network administrators can manage servers and perform
other management tasks from any computer that has a Web browser. While
this is a NetWare 5.1 tool, partial functionality is provided for NetWare 4.x
servers and above within the same directory tree.
The Management Portal will allow an administrator to perform the
following tasks:


Modify and view certain configuration parameters.

Load and unload NetWare Loadable Modules (NLMs).

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

380

Chapter 9

Network Operating Systems

NLMs are Novells way of customizing the server to run just the services you
want. You can load or unload NLMs to provide or remove different types of
services, like DHCP, DNS, FTP, or many others.

Start and stop certain processes.

Configure the memory system.

Check the way server memory is being used.

View and change registry settings.

View and clear server connections.

Set parameters for network interface card (NIC) drivers and disk drivers.

View the status of certain processes.

Manage disk volume information.

Compress large files.

Change the attributes of volumes and files.

Manage the file system, including changing the file system rights.

In addition to being able to manage your file server, you can also check on
the health status of the server. Take a look at Figure 9.2.
FIGURE 9.2

NetWare Management Portal

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Novell NetWare

381

As you can see, I am running a pretty healthy serverat least, everything

looks good at this point. I can get more information on the way the server is
operating by clicking one of the counters. For example, if I decided to look
at Packet Receive Buffers, the screen would look like Figure 9.3.
FIGURE 9.3

Packet Receive Buffers

Packet Receive Buffers are the area of memory that Novell uses to temporarily
hold incoming packets until the processor can process the information.

One of the big marketing points of all the NOS contenders is the use of IP.

NetWare 5.1 and IP

Ever since NetWare 5, Novell has been saying that their networks can run pure
Internet Protocol (IP). Pure IP is kind of misnomer, since it is pure only in the
sense that there is no Internetwork Packet Exchange (IPX) encapsulation or
any NetBIOS encapsulation that could be present in an NT environment. Since
the network can be configured so there is nothing but IP on the wire, you will
be eliminating network traffic devoted to those other protocols, and thus
saving bandwidth.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

382

Chapter 9

Network Operating Systems

Just because the folks at Novell keep harping on the pure IP bandwagon,
it does not mean that NetWare 5.1 is not backwardly compatible to IPX
environments. It is backwardly compatible, and the network administrator
can control when each of the other protocols comes off the network.

NDS eDirectory
When Novell introduced NDS, they harped on the fact that the directory service could be used to provide access and security to not only your network,
but to your partners and even your customers. Novell wanted third-party
vendors to jump on the NDS bandwagon, and that happened, but not to the
degree the people in Provo, Utah would have hoped.
The way Novell would like to see the eDirectory work is by allowing the
use of directory-enabled applications. With these applications, you could
customize the application so your customers can populate a customer-driven
directory. Your business partners can communicate directly with each other
and make use of each others directory.
An eDirectory can contain millions of objects in a single directory tree. It
offers Lightweight Directory Access Protocol (LDAP) v3 support, which is an
industry standard directory access protocol. You can administer eDirectory
using the Java-based administrative utility, ConsoleOne.

NetWare and Open Standards

NetWares move to pure IP means that many of the other open Internet
standards are supported natively. This includes things like these:


TCP/IP

LDAP

Extensible Markup Language (XML)

Structured Query Language (SQL)

Open Database Connectivity (ODBC)

Java Database Connectivity (JDBC)

Java Naming and Directory Interface (JNDI)

Secure Sockets Layer (SSL)

HTTP

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Novell NetWare

383

Workstation Management
For the last several years, Novell has provided a management piece for desktop management. With NetWare 5.1, the ZENworks Starter Pack is included
with the operating system. ZENworks includes the Novell Application
Launcher for publishing applications to the desktop as well as Workstation
Manager for desktop management. The ZENworks Starter Pack includes
applications that will allow the administrator to perform these tasks:


Distribute and maintain applications.

Automatically launch applications on the workstation.

Import information about the workstation into NDS eDirectory.

Distribute and update print drivers.

Centralize the location and administration of user profiles.

Check and possibly update client configurations without visiting the

workstations.

Automate software updates.

Create scheduled actions for users.

ConsoleOne
With NetWare 5, Novell introduced a new Java-based administration utility
called ConsoleOne. ConsoleOne allows an administrator to do these things:


Configure LDAP Services.

Manage NDS eDirectory objects.

Extend the eDirectory schema.

Set up user accounts.

Control eDirectory rights inheritance.

Manage NetWare file services.

Because ConsoleOne is Java based, organizations can create and establish

management from anywhere, as well as simplify development of customized
snap-ins. Figure 9.4 is a snapshot of the ConsoleOne interface.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

384

Chapter 9

Network Operating Systems

FIGURE 9.4

ConsoleOne interface

Web Collaboration and WebDAV Support

Web-based Distributed Authoring and Versioning (WebDAV) support is
also included with NetWare 5.1. With WebDAV, a client can be enabled to
communicate with a Web server for collaboration techniques. You can
manipulate Web folders and files to personalize the Web environment. You
can then access these files from any Web browser. By storing the files on the
server, you will be able to access them from any workstation. This also lends
support for Microsoft Office 2000. With WebDAV and HTTP support, it is
very easy to edit Web pages.

Novell Storage Services, FTP, and NNS

With NetWare 5, the maximum amount of disk space that could be
addressed was increased over previous versions of NetWare using Novell
Storage Services (NSS). NSS is a 64-bit indexed storage system that can
handle up to 8 terabytes of storage. Novell has changed the way the disk
storage system uses memory, and this has improved the volume mount time
considerably. For example, as a demonstration, Novell has crashed and
recovered a volume of 1 terabyte with over a billion files in 10 seconds.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Novell NetWare

385

File Transfer Protocol Server

NetWare 5.1 comes with NetWare File Transfer Protocol (FTP) Server. If
the FTP Server is utilized, any FTP client can transfer files to and from
NetWare volumes once they authenticate into the NDS eDirectory tree.
Once authenticated, users can access other NetWare servers that may not
be running the FTP Service.

NetWare News Server

Have you ever wondered how all those news groups on the Internet are
hosted? NetWare 5.1 comes with NetWare News Server so that your servers can host electronic discussions and maintain newsgroups for publishing
to the Internet or to the intranet. NetWare News Server is integrated with
eDirectory, so you can define news groups, add new group members, and
install other collaboration products. Users can access the information with
a standard newsreader like Netscape Communicator or Microsoft Outlook
Express.

Other NetWare Features

Here is a breakdown of some of the other features that come with NetWare 5.1.
NetWare Web Search Server lets you automatically index all of the
pages on any public Web site and search them for a piece of information.
NetWare MultiMedia Server includes tools to provide enhanced audio
and video capabilities, network-based staff training, and interactive customer
support.
Novell Distributed Print Services (NDPS) provides bidirectional communications between network printers, users, and administrators.
With NDPS users can locate network printers and get information on
the printers capabilities and configuration.
WAN Traffic Manager is a tool that can be used to create policies that
control eDirectory replication traffic over WAN links. The administrator
can define traffic criteria, and the Traffic Manager will compare communication requests to the criteria. If they match, the traffic will either be
allowed to proceed or it will be delayed.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

386

Chapter 9

Network Operating Systems

Oracle8i is an Internet development and database deployment tool,

which comes with a GUI-based management piece that will let you manage enterprise-wide databases. This tool also works with the Novell Java
Virtual Machine (JVM) that can be used to develop Java-based tools.
Novell Certificate Server 2 provides public key cryptography and digital certificates. With Certificate Server, you can deploy a Certificate
Authority (CA) management domain within NDS eDirectory. The CA is
used to generate certificates and manage public keys. These are then used
to enable the certificate-based security services such as SSL security to use
with LDAP Server.
Hot Plug PCI is supported for Network Interface Cards.

Minimum Installation Requirements

Here are the minimum hardware and software installation requirements from
Novell for NetWare 5.1. These were found at www.novell.com/products/
netware/details.

Hardware Requirements


Server class PC with Pentium II or higher processor

A VGA or higher-resolution display adapter (SVGA recommended)

A DOS partition of at least 50MB, with 35MB of available space

Available disk space outside of the DOS partition:



Standard NetWare products and WebSphere Application Server

for NetWare: 1.3GB on volume SYS:

RAM


Standard NetWare Products: 128MB

WebSphere Application Server for NetWare: 256MB (512MB

recommended) in addition to the Standard NetWare Products

Oracle8i with WebDB: 128MB (256MB recommended) in addition to Standard NetWare Products

One or more network boards

An ISO-9660 CD-ROM drive

A PS/2 or serial mouse recommended

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Novell NetWare

387

Software Requirements


DOS 3.3 or later, but not the versions of DOS included with Windows
95/98/ME/NT or 2000

DOS CD Drivers

NetWare 5.1 Operating System CD

NetWare 5.1 License diskette

Novell Client for DOS and Windows 3.1 (Network Installation only)

IP address

IP address of DNS Server

Registered Domain Name

Network board and storage device properties

Table 5.1 is a breakdown of some of the features of NetWare.

TABLE 9.1

Features of NetWare 5.1

Feature

Provided?

Ability to edit configuration files

Yes

Web Server integrations

IBM WebSphere Application Server

3.0 for NetWare, Enterprise Edition

Ability to use automatic settings for

a high-traffic Web site

Yes

Remote administration:
browser-based

Yes

Server status and performance

monitoring: browser-based

Yes

Certificate Services

Yes

Certificate Server included

Yes

DNS Server included

Yes

FTP Server included

Yes

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

388

Chapter 9

Network Operating Systems

TABLE 9.1

Features of NetWare 5.1 (continued)

Feature

Provided?

Group Management included

Yes

Group Management of links

Yes

Directory Service

Yes

HTTP 2.1 support

Yes

IP v6 support

No

Load balancing

Yes

NNTP Server included

Yes

SMP support

Yes

SSL 3

Yes

Supports hot-patching of OS

Yes

Supports hot-swapping hardware

Yes

Supports Application Server

Yes

Virtual directories

Yes

Microsoft Windows NT 4/Microsoft

Windows 2000

omparing the features and benefits of Windows NT 4 and

Windows 2000 is like comparing apples and oranges. They are the same
types of things, but they are just different.
Windows NT 4 could be considered Microsofts first commercially successful venture into the world of networking. Sure, they had LanMan and
they had Windows NT 3.51, but it really wasnt until NT 4 was released that
the marketplace really started to jump on the Microsoft networking bandwagon. With each Service Pack that was released for 4, more things got fixed
and usually a couple of new features got added.
Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Microsoft Windows NT 4/Microsoft Windows 2000

389

NT 4 grew and matured, and its life expectancy exceeded everyones

expectations as delay after delay kept postponing the release of Windows
2000. When Windows 2000 was released, it was said to be built on NT technology, but it was definitely the new kid on the block. When you look over
the white papers that Microsoft released on the migration from NT to 2000,
it becomes clear that Microsoft would like you to get rid of the old and
implement the new as quickly as possible.

Windows 2000/NT Server Features and Benefits

Giving you the features and benefits of Windows 2000 Server is a daunting task,
simply because there are several versions. There is the plain Windows 2000
Small Business Server, Windows 2000 Server, Windows 2000 Advanced Server,
and the Windows 2000 Datacenter Server. They are all based on Windows
2000, but the functionality (read that cost, and hardware requirements) increase
as you go up the line. I will try to keep this comparison of NT and 2000 as
generic as possible.

Directory Services
Directory Services is the major difference between NT 4 and Windows 2000.
As a matter of fact, many of the other differences spring from the fact that
Windows can now use a directory service. A directory service is just a way
to keep track of all that stuff on your network. It is a way to manage it,
secure it, and make it available to your users.
Scalability
In Windows NT 4, object management was done at the domain level, and
Microsoft suggested that you have a maximum of 40,000 objects per domain.
With Windows 2000, you can support millions of objects per domain or partition. Since the directory indexes the data store, the retrieval of information
about any object in the directory is fast. This information can be replicated
between sites, and the system administrator can manage when and how that
replication takes place.
Windows 2000 uses a system called the Global Catalog to provide a single
view of the directory objects that may be stored in multiple domains. This
Global Catalog is updated simultaneously with other replication cycles to
make sure the lag time is kept to a minimum.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

390

Chapter 9

Network Operating Systems

LDAP Support
The Active Directory is based on the Lightweight Directory Access Protocol
(LDAP). By using LDAP, Windows 2000 can get access to all features of the
namespace and integrate that into DNS. This makes finding objects in the
directory faster and easier.
Security Services
With Windows NT, you had centralized control of user and group security.
Windows 2000 adds support for things like Smart Cards and Kerberos. The
Catalog can enforce object-level and attribute-level security. There are no
restrictions on the security groups that span domains or partitions.

Network Object Management Utilities

If you can have millions of objects in a partition, how do you manage all of
them? With NT 4, there were several GUI interfaces and wizards that simplified management. With Windows 2000, there is a tool called the
Microsoft Management Console (MMC). There are several default types of
MMC provided with Windows 2000, but if they dont suit your needs, you
can create your own using snap-ins.
There are still plenty of wizards available if you are not sure how to perform
a specific task.
Group Policies
With NT 4, you could manage users, computers, and their desktops by using
system policies and settings. These were defined using the System Policy Editor and you could even make some of the policies mandatory. System Policies
worked most of the time. There were several problems with policies. When
policies were applied to accounts, user profiles would tend to get larger and
larger and it would take longer and longer for users to be able to log on to
either their desktop or the network. Next, when policies were no long necessary and were removed, they did not clean up after themselves neatly.
There might be residuals left over in registries that would come back and
haunt the administrator later.
Windows 2000 changed all that by instituting Group Policies. Group Policies can be assigned to certain Active Directory container objects or to user
objects. Group Policies can be designed to install software, update operating
systems, manage user profiles, and lock down a users desktop.
Group Policies can get very, very granular and can be applied to groups
of users. Group Policies are managed using the Group Policy snap-in to the
MMC, shown in Figure 9.5.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Microsoft Windows NT 4/Microsoft Windows 2000

FIGURE 9.5

391

Group Policy snap-in to MMC

Networking and Communications

Even Bill Gates will admit that Microsoft was a little late to jump on the
Internet bandwagon, but once the company did, it did it in a really big way.
Windows 2000 has made some great strides in the world of enterprise networking and Internet support.
Until Windows 95 came around, trying to configure a workstation to use
TCP/IP was a challenge. Because of the limitations of DOS, trying to fit all that
functionality under the 640KB memory ceiling was a real pain. Windows 95
changed all that and made configuration of an IP stack very easy. That functionality was also available in Windows NT and is certainly prevalent in
Windows 2000. Earlier versions of Windows networking relied heavily on the
use of Windows Internet Name Service (WINS) and the NetBIOS protocol.
With Windows 2000, Microsoft is trying to remove the operating system from
the constraints of WINS and NetBIOS and use TCP/IP entirely. Windows
2000 makes heavy use of Domain Name Service (DNS) and of Dynamic Host
Configuration Protocol (DHCP) to configure networks on the fly. As a matter
of fact, Windows 2000 makes such heavy use of TCP/IP that if your workstation starts up and isnt assigned a static IP address and if there isnt a DHCP
server available, Windows 2000 will just routinely assign a private IP address
to the workstation so it has something to work with.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

392

Chapter 9

Network Operating Systems

Quality of Service
Microsoft has implemented a Quality of Service (QoS) feature to Windows 2000
that will manage an end-to-end delivery system for all IP traffic. This includes
things that the user will never see but that can help administrators to manage the
amount of bandwidth being utilized and help to improve the quality of the connection for high-bandwidth and/or latency-sensitive applications. QoS was not
available in Windows NT.
Support for ATM and Gigabit Ethernet
As the enterprise network matures, so does the technology that moves packets between locations. Windows 2000 provides native support for both
Asynchronous Transfer Mode (ATM) and Gigabit Ethernet. Windows NT
relied on third-party vendors to make that support possible.
Virtual Private Networking (VPN) Support
One of the most talked-about technologies is the ability to create a private
network using parts of a public network like the Internet. This is referred to
as a Virtual Private Network (VPN). This technology has been growing
steadily and Microsoft worked with other vendors to create the Point-toPoint Tunneling Protocol (PPTP) to make it happen. PPTP can be used in
both Windows NT and Windows 2000 to provide VPN connections
between servers and also from a remote client to a server. Windows 2000
added support for the industry standard IPSec and Layer 2 Tunneling Protocol (L2TP) to enhance VPN security.
Remote Access Servers
Prior to Windows NT, connecting a remote desktop to a network was a challenge. Remote Access Server (RAS) simplified the process immensely. RAS
gives end users the ability to directly dial into the network. RAS provides
both direct dial and VPN support from a common user interface. RAS can
support up to 256 concurrent connections.
Windows 2000 expands the number of VPN protocols and also provides
user management through policies. The policies can be applied based on
the type of connection, the user or the group connecting, the time of day,
and the type of encryption, to mention a few. These can be applied using
a grant or deny access method.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Microsoft Windows NT 4/Microsoft Windows 2000

393

Remote Authentication Dial In User Service Support

The Remote Authentication Dial In User Service (RADIUS) client support is
integrated into the dial-up networking client. This allows the authentication
into an ISP network or non-Windows direct dial networks. Windows 2000
has complete RADIUS functionality. Windows 2000 supports RADIUS
authentication and RADIUS accounting. The information is stored as part of
the Active Directory or it can be stored as part of a local database.
The Windows 2000 RADIUS client has a graphical user interface that
provides access to most of the attributes you will need to configure your connection to another Microsoft computer. In addition, there is a tab that will
let the user tweak the connection for any vendor-specific attributes.
RADIUS also uses policies to manage user access.
Network Address Translation
Network Address Translation (NAT) is a technology that hides enterprise IP
addressing from the public network. It translates the private internal IP address
to a public external IP address.
In Windows NT 4, this technology was available if you added Microsoft
Proxy Server. Windows 2000 Server includes complete address and port
translations that will support any IPv4 client. Windows 2000 also includes
NAT translation for PPTP-based VPN traffic.
Multi-Protocol Routing
Windows NT 4 supported many of the most common routing protocols, including the Routing Information Protocol (RIP) v2 for IP, Service Advertising Protocol (SAP) for Internetwork Packet Exchange (IPX), Open Shortest Path First
(OSPF), DNS Proxy services, DHCP Relay Agents, and the Bandwidth Allocation Protocol. Routing with NT 4 worked with LAN Networks, FDDI, Frame
Relay, X.25, ISDN, and T1 transport.
Windows 2000 added integration into Active Directory as well as management using a Microsoft Management Console (MMC). In addition,
Windows 2000 supports Internet Group Messaging Protocol (IGMP).
IGMP allows multiple clients to share an IP multicast session. This reduces
traffic if you are using slower connections.

Disk Storage and File Sharing

Servers are around to provide access to information, and that information is
in the form of some kind of file stored on a disk. Managing those files, and
making sure the disk subsystems are dynamic enough to handle expansion,
are some of the foundations of a good Network Operating System (NOS).

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

394

Chapter 9

Network Operating Systems

Volume Management
It was easy to extend the size of a volume on a Windows NT server. Adding
free space to a volume was not the problem; it was having to reboot the
server after you did it that caused administrators everywhere all sorts of
grief. With Windows 2000, you can add free space to a volume and not have
to reboot the computer.
Link Tracking
In Windows 2000, link tracking will keep track of shortcuts and links to files
that reside on an NTFS volume and for some reason or another have
changed location. This Distributed Link Tracking was not available in NT 4.
File Compression
File compression is available in both NT 4 and Windows 2000.
Disk Quotas
In Windows NT 4, if you wanted to make sure that every user on your network could use no more than 25MB of disk space, you had to look to thirdparty software. With Windows 2000, you can institute per-user disk quotas.
Distributed File System Support
The Distributed File System (Dfs) was an add-on service that was available
in NT 4. In Windows 2000, you can provide a single view of shares that are
actually housed on several servers. These shares can even reside on different
file systems, such as NTFS, Novell, and NFS.
Dfs shares can also be configured to provide fault tolerance and load
balancing.

Printing
Several major advances have been made in printing since the NT 4 days.
Since all the printers are published in the Active Directory, users can quickly
locate the most convenient printing resource. All the printers that are shared
in a domain are available in the directory. As far as the number of printers
that are supported, Windows 2000 has over 2,600 printer drivers on the
Installation CD.
Finally, Windows 2000 can use the Internet Printing Protocol (IPP),
which lets users print directly to a URL over an intranet or over the Internet.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Microsoft Windows NT 4/Microsoft Windows 2000

395

Managing the Desktop

One of the major issues of any enterprise-wide solution is how do you get
out to manage all those desktops? How can you protect data, install
applications, make sure users dont delete things they are not supposed to
delete, or even how do you manage to install the desktop operating system on hundreds of workstations in an efficient manner? Again, these are
some of the things that the folks at Microsoft looked at when they were
writing Windows 2000. There were ways of managing these things in
Windows NT, but the ways just were not as elegant as they could be.
Managing the Users Data
In NT 4, this was done using the user profiles. By setting up profiles, documents
that were stored on a local machine could be synchronized with documents
stored on the server. By making sure sensitive or important information was
moved up to the server, the system administrator ensured that the documents
were being backed up.
In Windows 2000, users can roam the network, and if they use a Windows 2000 Professionalbased machine, they will always have access to their
data, their applications, and their computer preferences. Not only can users
have network services wherever they go on the network, they can take some
of the resources with them offline if they are using a laptop. When the system
reconnects to the network, the information will automatically synchronize
with the information found on the network.
Installing and Maintaining Software
With Windows NT, this was an add-on. In order to install and maintain software from a central location, you had to be using Microsoft Systems Management Server (SMS), which was an additional expense.
That was another thing that changed with Windows 2000. Administrators can now use Group Policies to specify applications that will always be
available to a single user or to a group of users. The same types of policies
can be used to maintain an application where, if a user does something dumb
and deletes files, the files will be automatically reinstalled.
Software management doesnt end with the installation and repair of software. You also have to worry about how you are going to uninstall software
from the computer after its useful life is over. Polices can also be used to
remove software.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

396

Chapter 9

Network Operating Systems

Remote Installation Services

If you are a network administrator, you know how difficult it can be when
a new raft of desktops shows up on the shipping room floor. These machines
may come in without any operating system (OS) installed, or with the wrong
operating system installed, and it is up to you to make sure the right OS and
the right settings are installed on dozens of machines. This usually has to be
done by yesterday, just to make your life a little more stressful.
While Windows NT could not help you, Windows 2000 has a way of
remotely installing Windows 2000 Professional. If the desktops will take
advantage of a remote boot technology called PXE, a PC can connect to a
Windows 2000 server and install Professional. If the system does not use
PXE technology, a set of remote boot diskettes can also be used, or you can
install the system from CD-ROM.

Windows Security
Security has certainly come into the forefront as an issue in recent years.
With Windows NT 4, security was provided through the Security Configuration Editor that came with Service Pack 4. This security editor allowed the
administrator to automate some of the global and local settings, including
things like changing some registry settings, maintaining access controls on
files and registry keys, and even the security configuration of system services.
These settings were defined as a template and then rolled out to selected
computers on the network.
With Windows 2000, the same thing is done using Group Policies. It is
much easier to manage, and it is all contained in the Active Directory. To
take advantage of that, you have to be authenticated into the directory.
Authentication
To authenticate into a Windows NT 4 network, you could provide a user
name and password or authenticate using several different Web standards. It
was great for the time, but there were problems with cross-platform support.
With Windows 2000, you not only have the same type of support for several forms of authentication but you also have built-in support for Kerberos
v5, which can be used for cross-platform authentication. There can also be
multiple authentications managed for both the client and for the server.
Because Windows 2000 is no longer using the only Domain-based
authentication, where everything had to go through a Primary Domain
Controller (PDC) or a Backup Domain Controller (BDC), there is an
increase of performance because of the reduced stress on the server.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Microsoft Windows NT 4/Microsoft Windows 2000

397

Public Key Infrastructure

The use of public and private keys for security is referred to as the Public Key
Infrastructure (PKI). In Windows NT 4, this was available with the Option
Pack. Once the Option Pack was installed, administrators could set up a certificate server and issue certificates to users. These users could then use those
certificates to authenticate themselves with the built-in Internet Information
Server (IIS). Certificates could also be mapped to an NT 4 user account so that
access control could be managed using the NT Access Control Lists (ACLs).
Windows 2000 removes the need for the Option Pack. There is a builtin Public Key Certificate Server as well as support for various third-party
Certificate Authority (CA) services. Windows 2000 also has built-in support for SmartCards.
Encrypting File System
Laptop computers have long been the bane of network security specialists.
Executives would be running around the country carrying corporate sensitive information on a laptop, and the laptop would be lost or stolen. Corporate secrets could be lost. Windows NT did not have anything built in to
combat that, but with Windows 2000, there is the Encrypting File System
(EFS). EFS is built into the Windows 2000 version of NTFS so files and directories can be easily encrypted and difficult to attack. This service is transparent to the end user.

Other Technologies
This is far from a comprehensive list of the services that Windows 2000 or
Windows NT has to offer. For example, we have not even looked at Terminal Server, which was a separate product with NT and is now included with
Windows 2000.

Terminal Server gives your remote users the opportunity to connect to the
network and run network applications at the server.

While the list does not cover all the features of Windows 2000, it should
be sufficient to get you through the exam.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

398

Chapter 9

Network Operating Systems

The Directory and Hardware

Whether you are using Novells NDS eDirectory or Microsofts

Active Directory, the fact that you have a listing of all the stuff on your network is going to require some additional hardware resources. Novells
NDS has been around for years, so there are some pretty hard and fast
numbers of what works and what doesnt in the real world. At the time this
book was written, Windows 2000 had been out for less than a year, so facts
and figures were a little more difficult to come by. I did find some information you may find useful.

NDS Sizing
This information was taken from a July 2000 article in Novell App Notes by
Nancy McLain, a senior technical writer from Novells DeveloperNet University. The article was titled System Requirements for NDS eDirectory.
One of the advantages that NDS has over Active Directory is that it can
be ported to different operating systems, including UNIX and Windows NT.
As a matter of fact, NDS ran on NT long before Active Directory and
Windows 2000 were released.
In the App Notes article, McLain starts out by laying out the minimum
hardware requirements to run NetWare 5.1 and eDirectory. Unless you are
setting up a system in a lab, these requirements are meaningless. Not many
of us would set up a production server running a PC with a Pentium 200
MHz processor and 64MB of RAM. The operating system may load, but it
certainly wouldnt perform to reasonable levels.
Lets start evaluating NDS by looking at the estimated disk space.

NDS Database Disk Space Requirements

NDS is part of a network operating system, and as with all network operating systems that we have talked about and will talk about, you can increase
performance by increasing the amount of server memory, disk space, and
processor speed.
NDS and NetWare
One of the limitations to the size of an NDS tree is the amount of disk space
you have available, as well as the disk input/output limitations. A typical
object in the NDS eDirectory uses 2KB to 5KB of disk space. That means you
can calculate the amount of disk space that you will require by multiplying

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

The Directory and Hardware

399

5KB by the number of objects in eDirectory. For example, 100 user objects
would take up 500KB of disk space.

2 to 5KB is an it depends number. The average object will be 2 to 5KB unless

it has a larger than normal number of attributes or perhaps contains data like
a picture, biometric information, or sounds. If that is the case, the number will
be larger.

NDS eDirectory and Windows NT and Linux

The AppNotes article put the size of the database running on NT or Linux
in a table, scaling the database from 100,000 objects, 1 million objects, and
100 million objects. It also measured the minimum hardware requirements
necessary to run NDS, the operating system, and other basic services. The
data is summarized in Table 9.2.
TABLE 9.2

NDS eDirectory Sizing

Number of
Objects

Minimum
Processor
Speed

Number of
Processors

Minimum
Hard Disk
Space

Minimum
Amount of
RAM

100,000

Pentium III
450

500MB

384MB

1,000,000

Pentium III
450

5GB

2GB

100,000,000

Pentium III
450

2 to 4

500GB

4GB

The suggestions in the table were based on 5MB per 1,000 users, at an
average object size of 5KB. As for memory, Novell suggests that you set aside
20% to 30% of the memory for the operating system and 70% to 80% of the
memory for the NDS database cache. If the server is running any other services, you should dedicate less memory to NDS and more memory to the services. You may need additional or more powerful processors if your server
is running additional services, including file and print, and handling a very
large number of authentications.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

400

Chapter 9

Network Operating Systems

Novell Recommended NDS Design Limits

Novells NDS design limits are contained in Table 9.3.
TABLE 9.3

NDS Design Limits

Partition Size

Limit

Total number of partitions in the tree

Unlimited

Number of child partitions per parent

150 partitions

Number of replicas per partition

50 replicas

Number of replicas stored on a

Replica Server

250 replicas

Microsoft Active Directory Sizing

In January 2000, just before Windows 2000 Server was released, a
Technical Information Document (TID) was released at http://
support.microsoft.com/support/kb/articles/Q222/0/19.ASP.
The TID was called DCPROMO Space Requirements for Active Directory Database and Log Files and was for all three versions of Windows
2000 Server.
As can be expected, the size of the Active Directory is dependent on the
number and kinds of objects that will be stored in the domain database.
In a minimum upgrade scenario, the Domain Controller Promotion
(DCPROMO) will allocate 200MB for the NTDS.DIT database, and 50MB
for the log files, or a total of 250MB if they are placed on the same hard
drive. As always, more free disk space is suggested.
If you are planning to upgrade an NT 4 server to Windows 2000, keep the
following things in mind:


Active Directory databases are larger than the Security Account Manager (SAM) file on an NT 4 domain controller with the same number
of objects.

The Active Directory database will continue to grow until deleted

entries are tombstoned every 60 days.

For maximum performance, the Active Directory database, the Active

Directory log files, and the Windows 2000 operating system should be
placed on separate physical hard drives.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

UNIX/Linux

401

The SYSVOL should be on a different physical partition than the operating system, the Active Directory database, and the Active Directory
log files. This will prevent the contents of the staging files from consuming all the available disk space.

Finally, To be safe, at least double your estimated size.

Next we will look at an operating system that doesnt feature a directory

like Active Directory or NDS eDirectory. Lets look at UNIX/Linux.

UNIX/Linux

f highlighting the features of Windows 2000 was a chore because of the

different versions of the operating system, this is impossibility. There are
dozens of different versions of both UNIX and Linux. To make matters
worse, since this is an open source code operating system, many installations
have tweaked things to make standardization impossible! Between you and
me, though, we will give this a shot.
Now up until this time, we have pretty much kept the discussion of
Operating System as Religion out of this. Up until this time. UNIX people
are pretty fanatical when it comes to support of their favorite operating system, and there is pretty much only one way of doing thingstheir way or the
highway. In many areas, they do have a point. UNIX has been around for
almost 30 years. That maturity brings a stable, mature high-end operating
system available for servers and supercomputers, while there are versions of
Linux available for workstations and some lower-end servers.

Network Compatibility
Whenever you start comparing UNIX/Linux to any of the competition, the topic
of proprietary protocols comes up almost immediately. One of the things that
UNIX and Linux systems are famous for is the interoperability they offer based
on what some people have called the universal technical standards and protocols. And they have a valid point. When you compare UNIX/Linux with Windows 2000 and Windows NT and NetWare, there are some major differences in
the way that they handle protocols. I dont know that I would go so far as to say
that Microsoft or Novell does not handle protocols according to standards,
because there are so many standards! Besides, if you look closely at the way

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

402

Chapter 9

Network Operating Systems

many of the standards are written, you will see that the standards writers made
room for definable options in many of the specifications. That means, if
Microsoft has gone in and changed the TCP/IP protocol to meet its specifications, it is certainly within its rights. If you have ever taken the TCP/IP class that
was associated with the Windows NT 4 track, you will notice that TCP/IP is usually referred to as Microsoft TCP/IP.
Lets be honest about a few other things here too. Microsoft has the marketing machine down to a science, and sometimes it appears that they will
expect the computing public to believe anything they say, just because it is
issued from Redmond. A case in point is the utilization of Kerberos v5. If you
read the Windows 2000 marketing materials, Kerberos v5 is touted as the next
best thing to sliced bread for cross-platform compatibility. Had the version
that was released with Windows 2000 been industry standard Kerberos v5, it
would have been. But it was another version of Microsofts Kerberos, and in
this case, Redmond, Washington did not even release the specifications until
after the official, on-the-shelves release of Windows 2000 Server. So, we will
stipulate here that Windows plays with standards, and UNIX tends to follow
the letter of the law more closely than other operating systems.
Following are some of the networking advantages that UNIX Servers offer.

Telnet
Administrators can Telnet into a remote host to perform routine administrative tasks. The administrator does not have to be sitting in front of the computer that requires the attention. This is also a cost savings, because a UNIX
machine can be operated headless, without a keyboard or a monitor. Telnet
is a two-edged sword. Information crossing the wire is sent in plain text format, so it can be captured. There are several alternatives to Telnet that
encrypt data across the wire.

Performance
UNIX and Linux provide faster read/write operations than other operating
systems. UNIX/Linux computers tend to operate for months or years without the need of a reboot. Crashes are rare. The number of reboots forced by
configuration changes are minimal in a UNIX/Linux environment. UNIX/
Linux are modular in nature, meaning that each can be adapted to changing
conditions without too much of a hassle. Since the boxes do tend to support
the high-end server platforms and run for years, most e-commerce sites tend
to run on UNIX boxes.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

UNIX/Linux

403

Linux especially looks strong when you compare the cost of the operating
system. Since the Linux operating system can be had for the price of a download, it is tough to beat in the cost comparison. In addition, the minimum
hardware requirements for a full-blown Linux server are such that some
desktop operating systems would not run on them.

Hardware
Because the hardware has a longer life in the UNIX world, more drivers tend
to be available and the network cards and other peripherals can be used for
a longer period of time. This minimizes the investment in hardware
upgrades. As a matter of fact, many UNIX/Linux implementations have to
stay with the older, more mature drivers because there are few new drivers
written specifically for UNIX and Linux. Since there are more implementations of other operating systems, these drivers get written more rapidly. That
trend is, in fact, changing. Linux is becoming more and more of a threat to
the other network operating systems, so hardware manufacturers are paying
attention and providing support earlier.
Some hardware manufacturers have really embraced UNIX and Linux.
For example, DEC Alpha, owned by Compaq, used to be touted as a Windows NT platform. Compaq has since turned away from NT or 2000 on the
Alpha platform. Meanwhile, there is a fully functional version of Linux for
the Alpha, as well as Compaqs own True64 UNIX.
Other platforms that have this type of support include Sparc, UltraSparc,
and the PowerPC, not to mention several others. Because in each case the
operating system requires little or no architectural change, the look and feel
of the operating system is similar regardless of the hardware platform.
IBM has used UNIX and Linux on its mainframe computers. As a matter
of fact, Linux machines can operate side by side, meaning that there are
scores of Linux operating systems running simultaneously on the same
machine or as virtual machines supported by another OS.

Automating Processes
UNIX/Linux administrators are able to automate many of the processes of
the operating system and of the applications by making use of shell scripting.
A shell script is a program that the administrator can write to automate certain administrative tasks. Meanwhile, on the Microsoft side of the house,
you might be using a graphical user interface wizard to perform the same
task. And theres the rub.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

404

Chapter 9

Network Operating Systems

Administrating a UNIX box is not for the faint of heart. It is a combination of some GUI interfaces that run on Java or of scripting and manually
configuring script and configuration files. Yes, the scripting and configuration files will lead to less overhead and faster performance. But, and this is
a big but, it also leads to human errors and administrative overhead. A good
UNIX administrator can fly through the administrative processes, but turning a new hire with some computer experience into a UNIX administrator is
going to require a sizable investment in time, if not capital. It is easier to turn
out a trained Windows 2000 administrator in far less time.
Now, like it or not, we are moving away from the command line world.
It pains me to say this, but there are hundreds of system administrators in the
field right now who could not write a batch file to save their lives. Or even
know what one does.

Security
It is difficult for me to compare the security of a UNIX Server to that of a
Microsoft Server or NetWare Server. All of these servers are as secure as their
administrators can make them. Each of the operating systems has security
flaws. Each of the operating systems has patches or workarounds that fix
those flaws. Some determined hacker with too much power and too much
time on his hands could access all of the operating systems. How many of the
holes are user caused, rather than operating system caused? It would be my
guess that security risks resulting from poorly configured or nonexistent firewalls outnumber the risks from operating system security holes.
As far as users writing programs that take advantage of the security holes
of each of the operating systems, both UNIX and Windows are victims of
their own open nature. Microsoft products are vulnerable to attack from
Visual Basic code. UNIX systems have a long list of known vulnerabilities.
Each has been victimized several times. Microsoft products tend to get hit
more often because there are more installations and the installations tend to
follow a certain pattern. For example, if you are hacking a Microsoft network, you stand a good chance that Exchange and Outlook will be there. In
addition, because it is a GUI-based operating system, you stand a better
chance of hitting a network with a more inexperienced administrator. All in
all, if you are going to write a virus to hit the world hard, hit Microsoft products. You stand a better chance of making an inroad.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

OS/2

405

Scalability
Now, here, the hands-down winner is UNIX/Linux. For years platforms like
the IBM RS6000, HPs 9000, Suns Ultra, and DEC Alphas have all tuned
the 32-bit and 64-bit UNIX operating systems to take full advantage of multiprocessor computing. Mainstream UNIX systems can take advantage of
multiple mainboards that can handle from 2 to over 100 processors. High
availability of these systems makes for almost 100% uptime.

OS/2

hile OS/2 is not a major player in the networking wars, it does have
its place. In the world of OS/2, the server operating system is referred to as
Warp Server.

Warp Server
Warp Server is designed to support large, multiserver networks that consist
of hundreds of users, as well as smaller departmental installations.

Supported File Systems

Warp Server features the Journaled File System (JFS), which is targeted for
application servers that will be using application middleware like Lotus
Notes, DB/2, Java Server, or Web Server. IBM says that the performance of
these application servers is comparable to the earlier 386-based HPFS-based
Warp Servers, but the new system supports multiple processors. JFS expands
the capacity limits of disk space, supporting partitions of up to 2 terabytes.
Files can also be 2 terabytes in size. The recovery time has also been
improved since the old HPFS file system.

TCP/IP
The Warp Server version of TCP/IP can support up to 64K concurrent socket
connections, which means there will be greater support for larger application
servers. Warp Server has improved its buffer management, it has been optimized to HTTP connections, and FTP and TFTP utilities have been rewritten
as multithreaded applications.

Administration
Administration is handled through a Graphical User Interface.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

406

Chapter 9

Network Operating Systems

Real World Scenario

It really pays to know your operating systems! Each of the operating systems I have described in this chapter brings special qualities to your network. Each operating system has some things that is does very well. The
same operating system may also have some tasks that it doesnt do so
well. Knowing these strengths and weaknesses will help you as your plan
your server implementations.
As much as the marketing people would love to have us believe that there are
Windows 2000 networks, or NetWare networks, or UNIX networks, the fact
remains that there are just networks. If the network is any size at all, chances
are you will have a smattering of at least two, if not three or four, operating
systems. It just makes sense. If I need to manage dial-in capabilities to a network, Microsoft networking products do a much better job of making the configuration and management easy than do other vendors. I would opt for a
Microsoft solution. In other areas, say file and print services, NetWare may be
the better choice. In each case, I am going to try to use the best solution for
the business case, regardless of the manufacturer.
This theory, though, puts a lot of pressure on the IT department. Now the
department not only has to know something about NetWare, but something
about NT/2000 and something about UNIX. It presents the department with
another set of unique challenges!

Installing a Network Operating System

nstalling and working with a Network Operating System (NOS) turns

out to be a completely different ball game than dealing with regular PC operating systems. There are a lot more details to think about. For example,
when installing a PC operating system such as Microsoft Windows in an
office environment, you certainly do have to consider network connectivity,
but you dont have the network concerns that you have with an NOS. For
example, with an NOS, you have to be very concerned about making sure
the computers NetBIOS name or its Domain Name System (DNS) Fully
Qualified Domain Name (FQDN) is registered with a name server and is
operational (i.e., that you can ping the host by name).

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Installing a Network Operating System

407

Servers can be equipped with high-availability and highly fault tolerant gear
such as RAID array adapters, data vaults, and other protective gear. An NOS
must be able to support the gear thats included with the computer for fault
tolerance purposes. Regular PC operating systems rarely utilize such fault tolerant measures.
Also, with an NOS you might be concerned about a list of users and their
associated attributes and the managing of the user database. In some NOS
instances the user database will be of a hierarchical natureNovell NetWare 4
and higher and Windows 2000 Active Directory (AD) use object-based hierarchical database trees for their user database. Typically PC operating systems
dont have a user database or, if they do, as is the case with Windows NT Workstation and Windows 2000 Professional, its seldom accessed or managed by
end users.
Installing an NOS can be as easy and carefree as a wizard-based installation
of Windows 2000 Advanced Server, or very complicated, requiring that you
know precisely what youre doing, as in the case of a UNIX computer running
HP/UX, IBMs AIX, or some other flavor of UNIX. But heres the deal:
Regardless of how easy or complicated the NOS installation is, you must not
take lightly the power of the NOS and overestimate your abilities to manage
its installation. This has been a problem for Microsoft since NTs inception.
Administrators (or, more appropriately, would-be administrators) think that
NT is so easy to install that a trained chimp could do it. And theyre right. But
the installation of the NOS isnt where the actual administration comes in.
Administration lies in the ability to professionally manage the server farm in
such a way as to provide a cohesive, intelligent work environment for your
user community. Installing the NOS is only a trivial percentage of the battle.
That being said, I would urge you to study to show yourself approved when
it comes to installing an NOS. How do you do this? You go and take a formal
class or set of classes from a reputable instructor on the installation, care, and
feeding of an NOS. In the Microsoft world (the one Im most comfortable
with) the classes are each a week long and youre required to take five or six
of them to begin to be really strong in Windows 2000 Advanced Server! Heres
an operating system thats Plug-and-Play and uses wizards to guide you
through the entire installation. What could possibly take six weeks to learn
about? Well, as Ive hinted above, there are hundreds of details that need to be
considered when working with highly advanced systems such as UNIX, NetWare, Linux, or Microsoft offerings. If youre a newbie to the industry, it is
important to take formal training in the server NOS of your choice and to
become literate in it. If youre familiar with one NOS but want to learn
another, the work isnt quite as hard because you already understand the

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

408

Chapter 9

Network Operating Systems

underlying principles of things such as network connectivity and user account

databases, but you still should take some training to master the nuances of the
system. This is especially true with Windows 2000, a different creature than
weve heretofore seen.
All NOS installations have basic components that youll be worried about
and can prepare for ahead of time. Heres a short list:


Youll need a CD player in the computer. Most NOS installations

today utilize CDs for their installation data.

Youll also need the latest installation CDs (preferably adequately

paid for and licensed).

Youll need to have at least one NIC in the computer and it should
already be connected to the network.

The computer youre installing to should be compatible with the NOS

you want to install (or vice-versa, depending on how you want to
think about it). For example, you probably cant install IBMs AIX
UNIX software on an HP K-class UNIX server. Wont work. Nor can
you install what some lovingly call Hockey Pucks (HP-UX) on an IBM
R6000.

TCP/IP addressing for the NOS should already be worked out. Most
servers utilize a hardwired IP address and subsequent TCP/IP configuration information.

Fault tolerance or other peripheral gear needs to be compatible with

the NOS youre installing.

You should be thinking about backups for the server ahead of time. If
a special module for your backup software is required to back up this
particular NOS, then you should have already obtained that module
and become familiar with it.

These are just a few of the things to think about when beginning your
installation of the NOS.

Installing NOS Updates

eriodically, service patches, service releases, service packswhatever

you want to call themare released for a given NOS. Sometimes the service
patches can be pretty involved to install. For example, installing a service

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Installing NOS Updates

409

pack on Microsofts Systems Management Server (SMSa piece of software

that isnt an NOS but runs on Microsofts NOS products) is a pretty tricky
operation. There was even a beta test for the SMS service pack prior to its
release! With Novell NetWare you may have to update various NetWare
Loadable Modules (NLMs) or Virtual Loadable Modules (VLMs) in order
to get the service patch installedwhich means youll have to download the
required NLM or VLM updates as well as the service patch itself.
You should be cognizant of new service patches that come out for your
NOS. This means that youll regularly visit the manufacturers Web site to
ascertain whats new relevant to your NOS. Some manufacturers offer an
e-mail service with updates to the NOSespecially security updates.
But that doesnt mean that you should immediately download and install
a new service patch when it comes out for your NOS. Many an admin has
gambled with this kind of thing and lost. Instead, test new service patches on
servers in your lab before putting them on production servers.
You should read the readme notes associated with a service patch to see
what kinds of things it patches. You may not need any of the functionality
that a specific service patch alleges to provide. On the other hand, sometimes
even though the service patch doesnt address anything specific to your particular servers operation, the code is updated and seems to cure other ills
that arent described in the readme. Am I describing a hit or miss process
with service patches? Let me give you a general rule of thumb: After a service
patch comes out, wait a few weeks before implementing and talk it over with
your server buddies. See if you can get the word on the street about how the
patch is running for those daring individuals whove gone out and immediately applied it. If you hear of success, then test in a lab and go forward. If
not, then take a wait-and-see approach.
You should be aware that sometimes youll be advised by a support technician of a specific patch or even a file that you need to download and apply in
order to (hopefully) solve a problem youre having. Ive had mixed luck with this
kind of thingbut Ive always found that technical support technicians wont
go much further with you if you dont take the time to apply the service patches
or files or whatever that theyre telling you to apply.
Note that some design specifications may require a specific NOS version
with specific service patches or service packs applied. This is not uncommon
at all. For example, you might run into a scenario where youre installing some
Web servers and the requirement for the Web application software youll be
installing later on down the road requires that you run Windows NT 4 with
Service Pack 6 (SP6) installed. Your own design specifications and comfort
level with various network operating systems running a given service patch or
service pack level may dictate that certain levels be put into place and tested
before putting the server into production.
Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

410

Chapter 9

Network Operating Systems

Updating Manufacturer-Specific Drivers

When installing gear on a server, either before or after the NOS

installation, its key that you download and apply the latest device drivers
for that device. This is especially key when brand new versions of an NOS
come out. For example, lets say that youre trying to install an HP CD
Writer on a Windows 2000 Advanced Server installation. But youve got
the Windows NT 4 device driver for that device. Its incumbent upon you
to go to the HP site and see if you can find a Windows 2000compatible
driver for the device instead of using old code (that probably wont work
correctly).
Sometimes the NOS manufacturer supplies a default device driver for common well-known gear. For example, 3COM NICs are so widely used and so
well-known that NOS manufacturers will have likely coded a device driver for
the majority of 3Com NICs in the world. But should you use the NOS native
device driver or should you go out and download the NIC manufacturers
driver? In most cases, I would advise you to download the manufacturers
device driver and use it in place of the driver that came with the NOS. Typically youll get more functionality out of a driver that was written by the company that manufactured the device. However, I have had instances where a
device driver that I downloaded from a manufacturer didnt work but the
NOS driver did. So again, its a hit-and-miss process. But Id err on the side of
sticking with manufacturer device drivers instead of those that natively ship
with the NOS.
Windows 2000 is so Plug-and-Play, by the way, that you might not need
to concern yourself with a device driver unless youve got some weird device
that isnt well-known and probably didnt have a device driver written by
Microsoft for it.

Installing Service Tools

s you go through an NOS installation, youll find that there are

options to the NOS that you might want to consider selecting. Typically you
can select these options at installation time or later, after the NOS is
installed.
Simple Network Management Protocol (SNMP), for example, is a service
that you might want to install on your servers. Lets say that youve got an
enterprise management tool such as HP OpenView (OV) and you want to

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Performing a Server Baseline

411

trap information from all of your servers into the OV product. You would
need SNMP installed on the server in order to accomplish this.
Network monitoring tools and agents are another add-on feature that is
often selected at installation time. With network monitoring agents, a network
monitoring tool (such as OV or Windows 2000 Network Monitor) can view
incoming and outgoing traffic on a given server and allow you to perform an
analysis of the traffic.
Some stuff comes automatically installed. Event logs, for example,
might be something that is installed by default and isnt an option you can
choose to install or not to install. On the other hand, perhaps you might
have to opt to install the logging service with a given NOS.
You should be aware that after the NOS has been installed, as you begin to
install application software you may find that the app installation installs specific NOS-based service tools that you had originally decided not to install. For
example, if youre installing HP OpenView Network Node Manager (NNM)
on a Windows NT 4 computer, the installation may very well detect that the
SNMP service isnt installed and will go through a process of installing it and
then allowing you to configure it. Less intelligent application software installations might simply put up a warning box telling you that a given service or
deamon isnt running and will then halt installation. Its up to you to decipher
what the error means.

Performing a Server Baseline

server baseline is a wonderful thing to perform right after you get the
NOS installed. If its 3:00 A.M. and youve been tussling with an NOS installation since 5:00 P.M. the previous evening, Im not advocating that you turn
right around and immediately obtain a baseline as soon as you get the NOS
installed, but it would be wise to baseline the installation in fairly short order
after installation.
Whats a baseline? Basically its a snapshot of the servers operational
characteristics as its running in its pristine state. Measurements such as disk
and CPU I/O, along with incoming and outgoing network activity, can paint
for you a picture of what the server acts like when it isnt in production and
isnt heavily loaded. Then, as you put it into production, you obtain another
snapshot. And as you add applications, you obtain another, and so on.
By performing some baselining of your server, you can really get a feel for
its performance health. For example, suppose that you baseline a server at its

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

412

Chapter 9

Network Operating Systems

inception and find that it runs wonderfully well. But then you install an
application and find that the CPU is hammered at 80100% most of the
time. Clearly you know that the application you just installed doesnt play
well in the sandbox with this server.
You can obtain third-party baselining software, but most NOS offerings
come with some form of server baselining capability, so you shouldnt have
to resort to add-on alternatives.

Adding Processors

dding processors to a system can be a scary thing if youve never done

it before. After all, youre essentially doing brain surgery. The following
chapter sections detail some informational items to be concerned with as you
go forward with your upgrade plans. I should point out that an upgrade
checklist is a wonderful thing to work up in preparing for a CPU upgrade.
You can add bullet points in there about checking the Web and FAQs for
important info relative to your upgrade, download the lasted documentation, schedule downtime for the upgrade, and so on. One of the wonderful
things that comes out of an upgrade checklist is the homework that you have
to go through to formulate your checklist. Just thinking through a project
causes you to ask questions and discover informational items that you may
not have considered before.

Performing an OS Upgrade to Support Multiprocessors

Before installing any upgrade, its a wise idea to sit down and think about the
steps youre going to go through, then double-check yourself, either through a
friend or peer or through system documentation. Reading the peripherals manual might provide wonderful insight into the steps youre supposed to go
through. Then again, maybe not. Ive read card documentation that was virtually no help at all. Its all up to the manufacturer to write a decent manual.
Your checklist should include the following items:


Make sure you note IRQs that are already in use with your system, the
base address used (if any), and any DMAs that are in use.

Write down the IRQ, base address and DMA, and other information
you intend to utilize with the new peripheral.

Note the peripherals serial and model numbers.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Upgrading System Monitoring Agents

413

Note the cables required; if they dont come with the peripheral,
obtain them.

Note any special software installation considerations.

Include a note to yourself to follow good ESD guidelines.

Make sure you perform a graceful shutdown on the server before commencing with installation.

Disconnect the computers power source(s).

Note any new or unusual behaviors. These may be situation-normal

with the new device, or they may signal a problem.

Note the screens that appear at power-on, especially looking for any
errors that might be displayed.

Installing server peripherals is easy to do, as long as you watch out for the
usual gotchas that surface. Unfortunately, these gotchas often happen at
2:00 A.M. when youre all by yourself in a server room and youre not thinking very clearly. Writing down a procedure checklist will be hugely helpful
to you in times such as this.

Upgrading System Monitoring Agents

here are three occasions when you might need to consider upgrading
system monitoring agents or when an agent is automatically upgraded. In two
of the situations youll automatically obtain upgrades. In the third situation,
youll have to consciously apply agent upgrades.
Installation of an upgrade NOS Some network operating systems come
with built-in system monitoring tools. Both Windows NT and 2000, for
example, come with such tools. In NT the program is called Performance
Monitor, and in Windows 2000 it is called System Monitor. If you
were to upgrade a Windows NT system to Windows 2000, Performance
Monitor would become System Monitor. Its functionality would be
largely the same, but some features might be addedto monitor Windows 2000 Kerberos events, as an example.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

414

Chapter 9

Network Operating Systems

Installation of a service pack or patch Upon installing a service pack or

patch, you might automatically gain upgrades to specific system monitoring agents. For example, perhaps an agent thats used to monitor hard
drive activity isnt working quite as anticipated. When a service patch is
released for the NOS, an upgrade to that specific agent might be applied
along with all of the other changes a patch is designed to handle.
Third-party system monitoring and proprietary agents If youre running a third-party system monitoring program such as NetIQ, for example, you may need to install proprietary agents to different clients
participating in the system. As new updates to the system monitoring
program come out, there will also probably be agent upgrades that
youll have to apply to the various clients. How you apply those
upgrades depends on the system involved. The point is this: Because the
system monitoring program isnt built into the NOS, youre responsible
for its care and feeding, whereas built-in components are often upgraded
by service patches or upgrade programs.
Its worth your while to prepare a preemptory checklist that you can use
when applying upgrades or patches that affect system monitoring programs.
For example, suppose that youre applying an NOS upgrade that you know
will impact built-in system monitoring features. Its to your benefit to
include on your checklist a reminder to do a post-installation check of the
system monitoring features to make sure theyre working and to see how
theyve changed relative to the old program.

Upgrading Service Tools

ervice tools fall into various categories, some of which have no standalone upgrade process that you might be able to go through, others of which
do have some sort of upgradability independent of their host NOS. Service
tools include diagnostic tools, EISA configuration utilities, system partition
diagnostics, and so forth.
Diagnostic tools such as Windows NT/2000s Network Monitor become
upgraded with OS upgrades or service packs that are applied. There is no
such thing as Network Monitorspecific upgrade. Diagnostic tools that are
part and parcel of an NOS are upgraded at NOS-upgrade time.
EISA configuration utilities are generally upgraded when you upgrade a
systems BIOS. There are some manually installable EISA configuration

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Creating a Baseline and Comparing Performance

415

utilities out there, but as a general rule of thumb, todays EISA comes with
the systems BIOS.
System partition diagnostic utilities such as FDISK and associated
Linux utilities are generally upgraded at NOS upgrade time, but can also
be individually upgraded or copied elsewhere. For example, you can copy
FDISK.EXE to a diskette and use it as a stand-alone partition configuration utility, independent of an OS or NOS. You probably dont want to
do that as a regular part of your operations, but Ive certainly seen it
done. You can also purchase third-party partition configuration utilities
such as Partition Magic that you can use on a take-it-with-me-to-thecomputer basis. The biggest caveat I have with this kind of technique is
that you need to be very cognizant of whether the utility will work on
your system. For example, you cant use an old copy of Partition Magic
on a Windows 2000 NTFS 5 partition. Youll irreversibly break it.
As with any other technique, its wise to prepare a checklist document
that you can use when considering an upgrade to diagnostic tools. In this
checklist youre aiming to describe what youve got now, where youre going
(or what youll have after upgrade), and how to validate that the upgrade
worked.

Creating a Baseline and Comparing

Performance

Something thats frequently overlooked by admins, but intensely

important, is the concept of obtaining a systems baseline. A baseline is a set
of operational characteristics that define how a server runs in its pristine
state, without additional applications or user load. Its important to have
data on how a system operates at this state because then youll be able to
diagnose problem spots when and if they surface. For example, if you know
that your 10,000 RPM SCSI disks operate at a certain level when theres no
load placed on them, if you suspect that client load is hammering the disks
you only have to compare your loaded information with your baseline info
to confirm your hunch.
Some network operating systems include the ability to gather baseline
information. Windows NT/2000, for example, can use Performance (System) Monitor to gather baseline statistics on various components and then
save the baseline to a file. You can then go back at any time and compare

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

416

Chapter 9

Network Operating Systems

current statistics with your baseline file to determine if you have a bottleneck somewhere in the system. With baselines, were primarily interested
in performance measurement. You can also save several different performance monitoring sessions, load them into a database, and query against
the results.
You can also buy performance monitoring software that will help you
create more robust views of your systems performance. Remember that a
baseline typically describes the systems performance in its pristine state.

Summary

In this chapter we have taken a look at the major network server operating systems. As always, before choosing, do your homework! Too many decisions are made by marketing hype or price. Do a study to find out what the
long term costs of running the operating systems are. My immediate suggestion to anyone who is getting ready to decide on an upgrade or a new installation would be to ignore all the marketing hype from all the vendors and do
your own homework. Go to user group meetings, read the news groups, take
a class, and listen to what the administrators have to say. In each and every
case, make up your own mind and dont be lured by the bells and whistles that
each of the operating system vendors will try and lure you with.
In this chapter I also talked about installing an NOS, pointing out that
while NOS software can be quite similar to desktop software, its usually
much more complex to install than a simple desktop OS installation. For
example, you must know networking information that goes beyond the ordinary client desktop connectivity with the network. Things such as name server
installations and correct statically entered IP configuration information are of
great importance in an NOS installation. Sizing the computer for the NOS is
important. Its highly advisable to obtain thorough formal product training
before going forward with your first production NOS installation.
I mentioned as well that software manufacturers periodically release
upgrades to their NOS software. These might come in the form of patches,
service packs, NLM and VLM updates, and so forth. Sometimes implementing a specific update requires that you update other system software
before you can go forward with the update youre interested in.
I talked about the wisdom of obtaining and applying vendor-specific
device drivers. Though some NOS software includes a plethora of prewritten

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Summary

417

device drivers, its usually wise to apply the driver that the devices manufacturer supplies.
I talked about performing an OS upgrade to support multiprocessors.
Network operating systems typically come ready to talk to a certain quantity
of multiprocessors and require an upgrade to the NOS, in the form of an
Original Equipment Manufacturer (OEM)supplied update or an actual
NOS upgrade produced by the NOS manufacturer. As always, an upgrade
checklist is an optimal way of assuring your success.
I also talked about the idea of a server baseline. You run performance
monitoring software, whether the software is bundled with the NOS or purchased from a separate manufacturer, in order to produce a baseline of the
systems operational characteristics at its new pristine state. Then, upon
loading of the server with applications, users, or other traffic, you can run
additional performance monitoring and compare it with your baseline to
make intelligent decisions about system performance and bottlenecks.
I mentioned that you can upgrade system monitoring agents, typically
through an NOS upgrade or the application of a service pack. When using a
third-party system-monitoring product you may have to manually update
client agents as new releases come out. Oftentimes the upgrading of a client
component is the first recommendation that comes from trouble support
desks and so you may be forced to comply with an agent upgrade in order to
determine if the upgrade rectifies a problem youre having. As always, a concise upgrade checklist is in order when working with such detailed items.
Along the lines of system monitoring agent upgrades, I also talked about
the need to upgrade service tools such as diagnostic tools, EISA configuration utilities, partition diagnostic tools, and so forth. Most of these upgrades
happen either through an NOS/OS upgrade or through a BIOS upgrade. You
may be able to perform an upgrade to a specific partition utility if its a single
file such as FDISK.EXE, though the need to do that is highly remote. As
always, a comprehensive upgrade checklist is in order.

Exam Essentials
Know basic information about Novell NetWare The NetWare installation starts with some files, including SERVER.EXE, being copied to a plain
old DOS partition. Once the operating system is installed, it will create data
storage volumes. If the file allocation tables of these volumes become corrupt, you will have to use a utility called VREPAIR to fix them.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

418

Chapter 9

Network Operating Systems

Know basic information about Microsoft Windows NT and Windows 2000

Windows NT and Windows 2000 can use either the old DOS file allocation
tables or the New Technology File System (NTFS). If a Windows machine
is having problems booting up, you will have to use an Emergency Repair
Disk (ERD) to fix the problem.
Know basic information about UNIX and LINUX UNIX and Linux
are used in high availability, high utilization servers. Configuration is
done through a series of text files, though some versions of Linux are now
providing more GUI utilities.
Know basic information about OS/2 OS/2 is very similar to Windows.
It is an IBM product.
Know how adding directory services or domains is going to impact your
server hardware Directory services and domains are databases that
track all the users, groups, servers, and peripherals on your network. They
are also used for security and authentication purposes. As with all databases, there is a lot of information being stored on these servers that will
directly impact the performance of your network. Be sure the servers are
capable of handling the information and have an availability rate that will
not make your life miserable.
Be able to install NOS Understand the concerns associated with the
installation and upgrade of an NOS.
Install device drivers Understand when and why youd install
manufacturer-specific device drivers.
Be able to install service tools Understand that you may have go back
and install specific service tools after youve performed your initial NOS
installation. SNMP, for example, might be something thats omitted at
NOS installation time, but is now needed for some function.
Be able to upgrade service tools Understand when and why youd
upgrade service tools such as diagnostic tools, EISA configuration utilities,
diagnostic partitions, and so forth. Understand that some components will
be upgraded at NOS upgrade time, while others may require a stand-alone
upgrade and yet others are upgraded at system BIOS upgrade time. Prepare
an upgrade checklist denoting your upgrade tasks.
Know how to perform a server baseline Understand why youd perform
a server baseline and what these baselines are useful for.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Summary

419

Be able to add processors Understand the concept of adding processors

to a server and the impact that the addition may have on the NOS. You
may have to apply an upgrade to the NOS for it to be able to recognize
and work with the additional processors. An upgrade checklist including
the reading of key information regarding the process is in order when
working with such detailed items.
Be able to upgrade system monitors Understand when and why youd
upgrade system monitoring agents and how that upgrade would occur.

Key Terms
Before you take the exam, be certain you are familiar with the following terms:
Access Control Lists (ACLs)
Asynchronous Transfer Mode (ATM)
Backup Domain Controller (BDC)
Bandwidth Allocation Protocol
Certificate Authority (CA)
DHCP Relay Agents
Domain Name Service (DNS)
Dynamic Host Configuration Protocol (DHCP)
Encrypting File System (EFS)
File Transfer Protocol (FTP)
Gigabit Ethernet
Group Policies
Internet Group Messaging Protocol (IGMP)
Internet Information Server (IIS)
Internet Protocol (IP)
Internetwork Packet Exchange (IPX)
IPSec
Java Virtual Machine (JVM)
Journaled File System (JFS)

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

420

Chapter 9

Network Operating Systems

Layer 2 Tunneling Protocol (L2TP)

Lightweight Directory Access Protocol (LDAP)
local area network (LAN)
Microsoft Management Console (MMC)
Microsoft Systems Management Server (SMS)
NetWare Directory Services (NDS)
NetWare Loadable Modules (NLMs)
Network Address Translation (NAT)
network operating system (NOS)
Novell Distributed Print Services (NDPS)
Novell Storage Services (NSS)
Open Shortest Path First (OSPF)
operating system (OS)
Packet Receive Buffers
Point-to-Point Tunneling Protocol (PPTP)
Primary Domain Controller (PDC)
Public Key Infrastructure (PKI)
Quality of Service (QoS)
Remote Access Server (RAS)
Remote Authentication Dial In User Service (RADIUS)
Routing Information Protocol (RIP)
Service Advertising Protocol (SAP)
shell scripting
SmartCard
Telnet
Virtual Private Network (VPN)
Warp Server
Web-based Distributed Authoring and Versioning (WebDAV)

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

421

Review Questions
1. Which network operating system is referred to as a Warp Server?
A. NetWare
B. UNIX/Linux
C. Windows 2000/NT
D. OS/2
2. Of all the network operating systems that we looked at in this section,

which one is the cheapest to acquire?

A. NetWare
B. UNIX
C. Windows 2000/NT
D. OS/2
E. Linux
3. According to this chapter, Novells eDirectory can be used on which

operating systems?
A. NetWare
B. UNIX/Linux
C. Windows NT
D. OS/2
E. All of the above
4. Which network operating system natively supports the JFS file system?
A. NetWare
B. UNIX/Linux
C. Windows 2000/NT
D. OS/2

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

422

Chapter 9

Network Operating Systems

5. Which network operating system (or systems) is not administered with

a graphical user interface?

A. NetWare
B. UNIX/Linux
C. Windows 2000/NT
D. OS/2
6. Red Hat has come out with an upgrade patch that fixes an issue with

an older Linux Loader (LILO) partition. What kind of upgrade will

you be performing? Select all answer that could be correct.
A. OS/NOS upgrade
B. Partition utility patch
C. Service patch
D. System monitoring tool patch
E. Service tool patch
7. Which network operating system has been used for super-servers

where stability and performance were mandatory?

A. NetWare
B. UNIX
C. Windows 2000/NT
D. OS/2
8. Which network directory service estimates the size of each object at 5KB?
A. Active Directory
B. LDAP
C. e-Directory
D. OS/2 Directory

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

423

9. Which of the network operating systems offers EFS?

A. NetWare
B. UNIX/Linux
C. Windows 2000
D. OS/2
10. What is Network Address Translation used for?
A. Replace DNS.
B. Replace DHCP.
C. Create a private network using very few registered IP addresses.
D. Create a firewall.
11. Which of the network operating systems mentioned can be used to

create Virtual Private Networks without any additional purchases?

A. NetWare
B. UNIX/Linux
C. Windows 2000
D. OS/2
12. Which directory service has been around the longest?
A. NDS
B. Active Directory
C. OS/2 Directory
D. None of the above

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

424

Chapter 9

Network Operating Systems

13. Red Hat has come out with an upgrade patch that fixes an issue with

an older Linux Loader (LILO) partition. What kind of upgrade will

you be performing? Select all answers that could be correct.
A. OS/NOS upgrade
B. Partition utility patch
C. Service patch
D. System monitoring tool patch
E. Service tool patch
14. Which operating system has been around the longest?
A. NetWare
B. UNIX/Linux
C. Windows 2000/NT
D. OS/2
15. Which NOS makes use of shell scripting?
A. NetWare
B. UNIX/Linux
C. Windows 2000/NT
D. OS/2
16. Which NOS would you most likely find using Visual Basic scripts?
A. NetWare
B. UNIX/Linux
C. Windows 2000/NT
D. OS/2

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

425

17. Which operating system would use files with the extension NCF?
A. NetWare
B. UNIX/Linux
C. Windows 2000/NT
D. OS/2
18. According to sources quoted in this chapter, what should be the

minimum size allocated for the Active Directory?

A. 125MB
B. 250MB
C. 500MB
D. 1GB
19. Once you have calculated about how much disk space Active Directory

should use, what does Microsoft recommend you do with that figure?
A. Implement it.
B. Halve it: Windows 2000 uses disk compression.
C. Double it.
D. Triple it.
20. Which of the following network operating systems use TCP/IP as their

preferred transport protocol?

A. NetWare
B. UNIX/Linux
C. Windows 2000/NT
D. OS/2
E. All of the above

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

426

Chapter 9

Network Operating Systems

Answers to Review Questions

1. D. OS/2 from IBM is known as the Warp Server.
2. E. Linux is available for the price of a download.
3. A, B, C. NDS eDirectory can be ported to UNIX/Linux and Windows NT,

as well as run on its native NetWare.

4. D. OS/2 natively supports JFS file system.
5. B. UNIX and Linux are primarily administered through script or

configuration files.
6. A, B, E. Were in a fuzzy gray area here simply because we dont

know how the vendor may decide to ship the patch. It could be that
a fix for LILO may require a complete OS/NOS upgrade. It could
also be that the upgrade simply means that we update a specific .EXE
file responsible for the maintenance of LILO. In either case, were
definitely applying a service tool patch.
7. B. As much as proponents of the other operating systems would love

to argue the point, UNIX is the operating system of choice for most
mission critical super-servers.
8. C. Novells eDirectory estimates the size of each object at 5KB.
9. C. Windows 2000 uses the Encrypting File System to provide added

protection to sensitive documents.

10. C. NAT is used to create a private network. In addition, it usually uses

just one (or very few) registered IP address.

11. C. Windows 2000 has VPN capability upon installation. With Net-

Ware, it would require the purchase of BorderManager.

12. A. Novells NDS was the first directory service to be used in the NOS

implementations mentioned in this chapter.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Answers to Review Questions

427

13. A, B, E. Were in a fuzzy gray area here simply because we dont

know how the vendor may decide to ship the patch. It could be that
a fix for LILO may require a complete OS/NOS upgrade. It could
also be that the upgrade simply means that we update a specific .EXE
file responsible for the maintenance of LILO. In either case, were
definitely applying a service tool patch.
14. B. UNIX has been around for over 30 years.
15. B. UNIX/Linux is configured using shell scripting.
16. C. Visual Basic is a Microsoft product and is an integral part of the

Windows 2000 and Windows NT environment.

17. A. NCF stands for Netware Control File, and it is used to batch com-

mands that will perform certain tasks.

18. B. The Microsoft Technical Information Document suggests a mini-

mum of 250MB.
19. C. Microsoft suggests you leave plenty of room of growth, so you

should double the amount of disk space allocated to Active Directory.

20. E. All of the network operating systems mentioned in this article

would like your network to run on nothing but TCP/IP.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Chapter

10

Network Interface
Technology
SERVER+ EXAM OBJECTIVES COVERED IN
THIS CHAPTER:

2.3 Install NOS


Configure network and verify network connectivity.

Verify network connectivity.

3.6 Upgrade adapters (e.g., NICs, SCSI cards, RAID, etc.).



Perform upgrade checklist including: locate and obtain latest

test drivers, OS updates, software, etc.; review FAQs,
instructions, facts and issues; test and pilot; schedule
downtime; implement using ESD best practices; confirm that
the upgrade has been recognized; review and baseline;
document the upgrade.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

ell, here we are, almost finished with the book, and

finally we get to talk about how the data packet gets put out on the wire, and
what happens to it once it gets there. In this chapter, we are going to look at
Ethernet and Token Ring and compare and contrast them. Once we finish
with the high-level overview, we will get down to the Network Interface
Card (NIC) level and see what we can do about describing how NICs relate
to the system bus. That discussion should be review if you spent time with
the chapter on bus architecture.
Lets start with the venerable, stable, easy-to-install Ethernet.

For complete coverage of objective 2.3, please also see Chapter 9. For complete
coverage of objective 3.6, please also see Chapters 1 and 2.

Ethernet, Then and Now

t is hard to imagine a world without Ethernet, isnt it? For most of us,
it has been around since we started in networking, and even though the specifications may have changed and the configurations may have changed somewhat, Ethernet is still what most networks are based on.
Between the invention of the PC and the start of the 21st century, there
have been many things that completely changed the way the world works.
One of those significant inventions came about in 1973 when Robert
Bob Metcalfe and D.R. Boggs were working at Xeroxs Palo Alto
Research Center (PARC), and they invented Ethernet.
Metcalfe started working at PARC in 1972, while he was working on
his Ph.D. from Harvard. If you are a student of computer history, you
know that PARC in the 1970s was where amazing things were developed.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Ethernet, Then and Now

431

It served as the birthplace of the PC, the graphical user interface (GUI),
and the page-description language (PDL). While all these monumental
discoveries were made in Palo Alto, Xerox somehow found a way not to
take advantage of them.
Ethernet is the Local Area Network (LAN) technology that allows us to
link PCs together into a cohesive unit. The amazing thing about this is that
the invention of Ethernet predated the invention of the personal computer,
yet this was the invention that would link more than 50 million PCs together
worldwide.
The original idea was published in 1976, when Metcalfe and Boggs
published a paper entitled, Ethernet: Distributed Packet Switching for
Local Computer Networks. Ethernet defined not only how the physical
media would connect, it defined how data could be transmitted across a
LAN at 10 megabits per second (Mbps). Figure 10.1 is a copy of one of
the figures included with the original patent for Ethernet.
FIGURE 10.1

Dr. Robert Metcalfes original patent figure of an Ethernet network

In 1976, Metcalfe shifted gears and went to the Xerox Systems Development
Division, where he managed the microprocessor and communication developments that formed the foundation for the early Xerox Star workstation. This
was the first workstation that included a bitmapped screen, a mouse, what-yousee-is-what-you-get (wysiwyg) word processing, Ethernet, and the software
necessary to include text and graphics in the same document.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

432

Chapter 10

Network Interface Technology

Metcalfe stayed at Xerox until 1979, when he left and founded 3Com
Corporation in Santa Clara, California. He formed 3Com to promote
Ethernet and PC LANs. Through the years, he was able to persuade Digital
Equipment, Intel, and Xerox to use Ethernet, which made it the most
widely used LAN.

Have you ever wondered what the three coms were in 3Com? They were
computer, communication, and compatibility.

Finally, the ownership of the Ethernet specifications transitioned from

Xerox to the Institute of Electrical and Electronic Engineers (IEEE). They
approved and released the Ethernet standard as IEEE Std. 802.3 in 1983. In
1985, the International Organization for Standardization (ISO) released the
first draft of the international standard as ISO/IEC 8802-3. This standard
established Ethernet as a true international force to be reckoned with.

Yeah, I know, this is one of those geek Mecca things. Would you like to see the original application for the patent of Ethernet? Check out http://164.195.100.11/
netacgi/nph-Parser?Sect1=PTO1&Sect2=HITOFF&d=PALL&p=1&u=/netahtml/
srchnum.htm&r=1&f=G&l=50&s1='4,063,220'.WKU.&OS=PN/4,063,220&RS=PN/
4,063,220.

Ethernet Basics
Ethernet uses CSMA/CD as a method of accessing the physical media. This
physical media also has some cryptic names like 10Base2, 10Base5, and
10Base-T. To complicate matters even further, there are different categories
of media, so you may be told to order several spools of Cat 5 10Base-T cable.
Lets see what we can do about translating some of this.

CSMA/CD
CSMA/CD stands for Carrier Sense Multiple Access Collision Detection. This
may be an oversimplified description of how Ethernet gets a packet out on the
wire, but it should make the point. When I was growing up, we used to spend
our summers in a small town in the Upper Peninsula of Michigan, Iron River.
My grandfather had purchased three lake lots and had built cabins on them for
his children, so every year we went up there and spent several weeks. It was a
great place to be a kid.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Ethernet, Then and Now

433

The adults didnt have it so bad either: There was no stress, no television,
one radio station, and one phone in my grandparents cabin that was hooked
up to a party line. Since there were not many year-round residents on the
lake, just about every cottage that had a phone was on a party line. They
came in several styles: There was semi-private, with two homes sharing the
same line. Then there were four-party and eight-party lines. We were on a
four-party line. Every time the phone would ring, we would all hold our
breath to see who it was for. If there were two short rings and one long ring,
someone was calling the Govanuses.
When we wanted to make a call, we couldnt just pick up the phone and
start dialing. We had to pick up the phone and listen (Carrier Sense) because
there could be someone else from one of the other parties using the phone
(Multiple Access). Once we had listened, and realized it was safe to dial, we
would start to dial the number. Once in a great while, someone else would
pick up the phone and start dialing at the same time, and both calls wouldnt
be completed (Collision Detection).
That is just the way Ethernet works. If a computer wants to send a packet, it
senses the carrier to see if there is traffic. If there isnt any traffic, the computer
begins to send the packet. If for some reason there is a collision with a packet
sent by another computer, both computers stop sending and wait for a short
period of time before trying to resend again.

Baseband vs Broadband
This may fall under the category of some basic electrical engineering, but
even if you already know this stuff, stick with me for a few minutes.
Any medium that can carry a signal has a certain capacity. This capacity is
referred to as bandwidth, and electrical engineers measure it in Hertz (Hz). Bandwidth literally stands for the width of a band of frequencies. To arrive at a bandwidth, you simply subtract the lower limit of the frequency from the upper limit
of the frequency. Nyquists theorem puts it much more scientifically: It relates the
bandwidth to the data rate as the data signal with a transmission rate of 2W,
which can be carried by frequency W, that is sufficient to carry the data. The converse of the theorem is also true: If you are given the bandwidth W, the highest signal rate is 2W. This data signal, if it is encoded in binary, has a data capacity in bits
per second (bps) that is twice the bandwidth in Hertz. This capacity can be
increased with multilevel signaling by transmitting more bits per data signal unit.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

434

Chapter 10

Network Interface Technology

The increase in capacity is the good news about multilevel signaling.

The bad news is that in order to use it, there must be a way to distinguish
between the signal and outside interference, which is called noise. You
may have heard or read specifications for electronic components that
mention the signal-to-noise ratio. This is not a theorem, it is Shannons
Law, and that sets an upper limit on the bps/Hz ratio, which will increase
logarithmically with the signal-to-noise ratio. In theory, you should be
able to get somewhere between 2bps/Hz and 12bps/Hz. That is in theory.
In reality, current technology gets between 1bps/Hz and 4bps/Hz. To
make matters simple, I am just going to refer to bandwidth as bps (bits
per second).

Actually, these are two different things. Bandwidth is a measure of the range
of frequencies used in an analog signal, while bits per second is a measure of
a digital data rate.

Lets use your telephone as an example. In a typical telephone, the

audio signals are limited to a range between 300Hz and 3,300Hz. A total
of 3kHz bandwidth is needed to transmit the signal. This signal can then
be encoded onto the bandwidth with about 10bps/Hz, and the data can
then be transmitted using a modem at 28.8kbps.

Multiple Access
So, the term band means a range of frequencies. If you decide to share a given
band, you can divide that band into smaller pieces, which are called channels.
These channels will then have just a fraction of the total available bandwidth.
As an example, lets say that you use the 9MHz wide band from 1MHz to
10MHz and you divide it into 3,000 channels, each with 3kHz of bandwidth.
This way, you could allow a great many telephone calls to be transmitted over
a single pair of wires, but you would have to manipulate the analog signal to
use just one channel.
There is another way to approach the problem. Suppose you just decided to
use the entire 9MHz as one signal carrier. This would eliminate the need to mess
with an analog signal to place a two-level digital signal on a wire pair. The tricky
part with this plan comes when you want to share the wire.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Ethernet, Then and Now

435

In RS-232 serial communications, one wire is used to send data one way,
and the second wire is used to send data the other way. The third wire is used
as a signal ground. This comes from one of the basic tenets of physics, in that
voltages have to be measured as the difference in electrical potential between
two wires. One of these wires is said to be the ground wire, while the other
wire is used to carry the signal.
Ethernet is a baseband system that places bits on a pair of wires using
Manchester encoding. Take a look at Figure 10.2 to see what a Manchester
encoding signal looks like.
FIGURE 10.2

Manchester encoding signal

Bit Boundary

Bit Boundary

Manchester encoding is a synchronous clock encoding technique. If you

have taken the Network+ exam, you are intimately familiar with the seven
layers of the OSI model. This is at the very bottom of the OSI model, working at the Physical Layer. Synchronous clocking means that the clock signal
is constantly present, so you can measure the absence or presence of voltage
on a line at a certain mutually agreed upon time interval.
In this method, binary data is sent over the wire, not a logical sequence of
1s and 0s. The bits are translated into a different format than straight
binary encoding. Table 10.1 shows the rules of Manchester encoding.
TABLE 10.1

Manchester Encoding Rules

Original Data

Value Sent

Logic 0

1 to 0 with a downward transition at

the center of the bit

Logic 1

0 to 1 with an upward transition at

the center of the bit

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

436

Chapter 10

Network Interface Technology

Lets break down Figure 10.2 and see what is actually happening. Figure 10.3
shows the first bit.
FIGURE 10.3

First bit in the Manchester encoding sample

Bit Boundary

Bit Boundary

In Manchester encoding, a 1 is indicated by the 0 to 1 upward transition

at the center of the bit, and a 0 is indicated by a 1 to 0 downward transition
at the center of the bit. Since the figure shows a downward transition at the
center of the bit, this must be from 1 (signal present) to 0 (signal not present)
or a logic 0.
The second bit in the example looks just like the first, so that must be
another 0. Look at the third bit, which is shown in Figure 10.4.
FIGURE 10.4

Third bit in the Manchester encoding sample

Bit Boundary

Copyright 2001 SYBEX, Inc., Alameda, CA

Bit Boundary

www.sybex.com

Ethernet, Then and Now

437

As you can see, the transition at the middle of the bit goes from 0 (signal
not present) to 1 (signal present). So, that means the data that was being
transmitted was a 1. So far we have 0, 0, 1.
Lets look at the last three bits in the diagram to see what we can come up
with. Figure 10.5 presents the bits.
FIGURE 10.5

Last three bits in the Manchester encoding sample

Bit Boundary

Bit Boundary

Bit Boundary

Bit Boundary

The first bit transitions from 1 to 0, so it is a 0. The next bit transitions

from 0 to 1, so it is a 1, and the final bit transitions from 0 to1, so it is also
a 1. That leaves us with a transfer of 0, 0, 1, 0, 1, 1.
If you looked at the diagram closely, you would have noticed that there
are not always signal translations at the bit boundary, but there is always a
signal transition in the middle of the bit.
Now, that didnt seem that hard, did it? In our small example it is a piece
of cake, but to achieve the Ethernet throughput of 10 megabits per second
(Mbps), there must be 20 million transactions per second. Now that presents
a challenge!

Ethernet Cabling
Ethernet cabling runs the size gamut from large and unwieldy to small and
easy to install. It all started with a cabling technology called 10Base5.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

438

Chapter 10

Network Interface Technology

Thicknet
Ethernet cabling, like the specification itself, has come a long way. The first
type of cable that was used was called 10Base5 by the IEEE. Everyone else
called it thicknet. The IEEE Specification was more descriptive, since it
referred to its main specifications. The 10 referred to 10Mbps, Base stood
for baseband, and the 5 referred to the 5 X 100 meter segments (for a total
usable distance of 500 meters) that could be used to link computers together.
If you had ever seen some thicknet, you would not have forgotten it. The
cable itself was about as big around as a United States dime. The flexibility
of the cable rivaled the dime as well: It was not very easy to work with. To
attach a node to the thicknet backbone, you used a drop cable from the node
to the backbone. To connect the drop cable to backbone, you used a device
called a vampire tap. The vampire tap and the rest of the configuration
looked like Figure 10.6.
FIGURE 10.6

10Base5 configuration

Thicknet

Vampire Taps

Transceiver

Transceiver
Cable

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Ethernet, Then and Now

439

The vampire tap had two sharp prongs, which cut through the sheathing
around the cable and made the electrical connection.
Thicknet was used in a bus topology. It could support up to 100 nodes per
backbone segment. Information was transmitted at 10Mbps and carried a
signal for 500 meters or about 1,640 feet. You could have a maximum of five
backbone segments connected using repeaters, according to the IEEE 802.3
specifications.

The maximum cable length did not include the length of the drop cables, just
the length of the backbone.

The length of thicknet cable between the drop cable connections had to be
at least 2.5 meters, or about 8 feet. Terminators had to be installed at each
end of the cable run. One end had to be grounded.
Thicknet was great at the time, but it was expensive, and because of the
weight of the cable and its lack of flexibility, it was a bear to install and
configure. Some of these issues were addressed with ThinNet.

5-4-3 Rule
As you read over the cabling specifications of Ethernet, you will notice
that the maximum number of backbone segments that can be connected
together seems to stay constant at five. There is a well known rule for
Ethernet implementations called the 5-4-3 rule. It states that, for Ethernet
to work, you can have five segments connected through four repeaters
with three of the segments being populated.

ThinNet
Next up came RG58A/U coaxial cable. When this was added to the Ethernet
specification, it was referred to as 10Base2. Like 10Base5, 10Base2 had
meaning. 10Mbps, baseband, and a single Ethernet segment could be 185
meters in length. This was also referred to as either ThinNet or, in deference
to the fiscally conservative, CheaperNet. In a ThinNet configuration, you
could have up to 30 devices linked together using British Naval Connector
(BNC) T-connectors on a single Ethernet segment. The minimum distance
between two nodes was set at 0.5 meters. You could also have up to five
10Base2 segments linked together, using four repeaters. The length of all the
segments could not exceed 925 meters.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

440

Chapter 10

Network Interface Technology

ThinNet network segments had to be terminated at each end, and one and
only one end should be grounded. A ThinNet network run was a Logical Bus
and Physical Bus. This means that each computer on the segment received
the electrical signal and each computer was linked to the next, as shown in
Figure 10.7.
FIGURE 10.7

Logical and Physical Bus

Workstation

Workstation

Workstation
Terminator
and Ground

Terminator

Workstation

Workstation

Twisted Pair
Probably the most common type of cabling in Ethernet networks is twisted
pair, otherwise known as 10Base-T. 10Base-T refers to unshielded twisted
pair (UTP) cable that was designated by the standards document ANSI/TIA/
EIA-658-A. This document designated different categories of UTP, including
Category 3, Category 4, and Category 5. Category 3 cable is 100 ohm cable
rated to 16MHz, Category 4 cable is 100 ohm cable rated to 20MHz, and
Category 5 cable is rated to 100MHz operation.
While the other cabling standards we looked at were a Logical Bus and a
Physical Bus, 10Base-T makes use of a Logical Bus and a Physical Star. Take
a look at Figure 10.8 to see what I mean.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Ethernet, Then and Now

FIGURE 10.8

441

Logical Bus, Physical Star Ethernet Network

Workstation

Workstation

Workstation

Hub

Workstation

Workstation

Workstation

In a Logical Bus and Physical Star, all the computers are hooked into a hub.
When a signal is sent from one computer to another, all computers get the signal,
just as if they were hooked together in a straight line. The computers are all
wired to the hub, so that is the single point of contact for all computers.
UTP, 10Base-T, is relatively easy to install, it is inexpensive, and it makes it easy
to reconfigure networks. The cable is susceptible to EMI as well as eavesdropping.

There are some restrictions. For example, devices may be located up to 100 meters
from the hub. Typically, the more devices you have on a particular segment, the
greater the chances of collision. The more collisions, the slower the network
appears to operateand the more your phone rings. See the section later on
switches versus hubs.

10Broad36
So far, we have been spending all of our time on baseband technology. We did
mention broadband earlier, also. 10Broad36 supports a cable type broadband
system to move data at a rate of 10Mbps. The 36 in the name is the distance
limitation, which in this case is 3,600 meters between any two stations.
10Broad36 makes use of the same coaxial cable that is used in cable TV
(CATV) systems. The cable is cheap and it is readily available. The broadband
cable system can handle support for multiple transmission services by dividing
the bandwidth into different frequencies, and then assigning a frequency to a
different service. This is the same way it is done in cable TV. This capability
means that 10Broad36 can share that cable with other services like video.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

442

Chapter 10

Network Interface Technology

You will notice that the maximum distance here is much longer than the
specifications for baseband coaxial cable used with 10Base5 and 10Base2.
Single segments of 10Broad36 can be as long as 1,800 meters. There are
some similarities to the baseband networks, though. Each of the 10Broad36
networks must be terminated by a head end device. This head end can be at
the end of a single 10Broad36 segment or at the start of multiple segments.
Nodes are attached to a 10Broad36 segment with a transceiver that is both
physically and electrically attached to the broadband cable. The computer
then attaches to the transceiver using an Attachment Unit Interface (AUI)
cable that can be 50 meters in length.
Broadband transmissions differ from baseband in the direction of the signal
flow. With baseband, signal flow is bidirectional, meaning it flows in both directions away from the transmitting node. Ethernet uses baseband; since baseband
can carry only one signal at a time, two pairs are required one for transmit and
the other for receive. With broadband transmission, the signal flows in just one
direction along the cable. That means that if the signal is going to reach all the
devices on the network, there have to be two paths for the data flow. This is done
with either single cable or a dual cable installation. With single cable, there are
transmissions occurring over two different channels, each channel using a different frequency range. One channel is used to transmit, and the other channel is used
to receive.
A dual channel configuration is just like the name sounds. Every station
is attached to two cables, one to send and one to receive. 10Broad36 cannot
support full duplex operationssimultaneous transmissions in both directions. Today, most Ethernet cards are capable of full duplex operation.
Up until now, we have dealt with the copper wire method of transferring
data. Lets take a look at some fiber optic implementations.

Fiber Optics
No matter what type of copper cable you use, there are disadvantages.
The signal degradation over long distance is one, and it is highly susceptible to electromagnetic interference. The advantage is that it is easy to
install and cheap.
As local area networks grew into campus area networks, metropolitan area
networks, and wide area networks, the need for longer distance communications
became greater. The solution came in the form of fiber optic cabling where the
electrical impulses were transformed to light and sent over glass or plastic cables.
This eliminated the problems of signal degradation due to distance, as well as
problems with EMI. As you can imagine, there are several different types of fiber
that can be used with Ethernet.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Ethernet, Then and Now

443

Fiber Optic Inter-Repeater Link

The fiber optic inter-repeater link (FOIRL) technology carries a 10 megabit
per second transmission rate that works with two cables. When the standard
was released in 1987, it was originally designed to be used over long distances
to connect two repeaters. The standard remained in effect until the 10Base
Fiber Link (10Base-FL) standard was released in 1993.
FOIRL can be used for point-to-point links up to 1,000 meters in length.
While the standard restricts implementations between two repeaters, some
vendors have devised ways to use FOIRL between a computer and a
repeater.

10Base-F
Another fiber standard is the 10Base-F, which also transmits data at
10Mbps over fiber optics. It was released as an enhancement to FIORL.
The identifier 10Base-F refers to not one, but three, fiber optic segment
types. These are 10Base-FL, 10Base-FB, and 10Base-FP. Each of these will
be discussed later.
Each of these interfaces is incompatible with the others.

10Base-FL
10Base-FL has already been mentioned in the discussion of FOIRL. Since it is
compatible with FOIRL, you can assume that some of the same characteristics
are shared. For example, both move data at 10Mbps over two fiber optic
cables. The difference between the two standards is in the distance over which
they can be used. 10Base-FL can be used over 1,000 meters.
10Base-FL can be used to connect two computers, two repeaters, or a
computer and a repeater. All connections are point-to-point through a
transceiver. If you are connecting a computer, the Network Interface Card
(NIC) will connect through an external transceiver through an AUI cable.
The transceiver then attaches to the two fiber cables through connectors
known as ST connectors. One cable transmits data and the other is used to
receive data.

ST connectors are officially named BFOC/2.5 in the standard.

10Base-FL is usually used for connections between buildings.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

444

Chapter 10

Network Interface Technology

10Base Fiber Backbone

10Base Fiber Backbone (10Base-FB) also moves data at 10Mbps, but it uses a
special synchronous signaling link that is designed for interconnecting repeaters.
This protocol allows more repeaters to be used in a 10Mbps Ethernet system.
Like 10Base-FL, these segments can be up to 2,000 meters in length.
The 5-4-3 rule was mentioned earlier. You are probably wondering why
it was an issue. Basically there were two factors that made it necessary to
limit the number of repeaters. First of all, repeaters add signal delay. Signal
delay can cause the maximum time for collision to move through the entire
network. Secondly, repeaters introduce a random bit loss in the preamble of
the packet. This can result in what is referred to as shrinkage in the gap
between frames. 10Base-FB combats these by solving the problem of the
interframe gap shrinkage by synchronizing the transmission between the two
repeaters.
10Base-FB can only be used in a point-to-point link between repeaters.

10Base Fiber Passive

10Base Fiber Passive (10Base-FP) also operates at 10Mbps in what is
referred to as a Passive Star system. The segments here are considerably
shorter, just up to 500 meters in length. 10Base-FP can be used to link up to
33 computers together, however.
The Star is a passive device. Passive devices do not require power, so it is
ideal for use in an area where an active power supply may not be available. The
Star acts like a hub, receiving the optical signals and distributing it uniformly
to all the transceivers connected to the Star.
10Base-FP is not capable of supporting full duplex modes of operations
and it has not been widely accepted.

Fast Ethernet

he Ethernet specification had data flowing at 10Mbps. When it was

first introduced, it was the fastest thing on the block. It certainly seemed like
a screamer compared to the old Arcnet implementations, which operated at
2.5Mbps. As the demands for network throughput increased, Ethernet
evolved into Fast Ethernet. Fast Ethernet pushes data at 100Mbps. It is backwardly compatible to Ethernet, with hosts running Fast Ethernet and Ethernet
fully able to coexist on the same network segment. Fast Ethernet is commonly
referred to as 100Base-T.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Fast Ethernet

445

Fast Ethernet is defined in the IEEE specification 802.3u.

Fast Ethernet uses the same frame structure, addressing scheme and
CSMA/CD access method as Ethernet. The difference is in the timing, which
must be scaled by a factor of 10 when configuring a Fast Ethernet network.
So, what does 100Mbps Ethernet buy you? Well, it depends on the types
of applications you are running and the size of file transfer you are doing. If
you are using some bandwidth hungry applications like video or audio transmission, you will notice a difference. If, on the other hand, you are using
small reads and writes, most of that data will be placed in small Ethernet
frames. The speed of this network is more likely to be impacted by things
other than the speed or bandwidth of the wire.
One of the ways that 100Mbps Ethernet shines over its slow predecessors is in
collision recovery. When there is a collision on 10Mbps Ethernet, the two systems
will back off for a set period of time before trying to resend their packets. This
backoff time is now 1/10th of what it would be with 10Mbps Ethernet. If you have
a crowded network, 100Mbs Ethernet may show dramatic improvements over
the 10Mbps Ethernet.

100Mbps Ethernet Physical Media

If you look at the 802.3u specifications, you will see that the physical media
that supports 100Mbps Ethernet has been named 100Base-TX and
100Base-FX. All of these media share a common Media Access Control
(MAC) specification, but when you get down to the Physical Layer, you will
see the transceivers are different. These transceivers can be part of a network
device like a repeater or a NIC, or can be external to the device. If it is an
external device, the transceiver is attached to the repeater or to the NIC using
a 40-pin Media Independent Interface (MII) connector. The transceiver is
then plugged into the MII connector or it can be attached with a MII cable
that is similar to the AUI cable that is used in the 10Mbps standard. These
MII cables have to be less than 0.5 meters in length.
The MII devices can support both the 100Mbps standard and the 10Mbps
standard. It does this with a process called auto-negotiation.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

446

Chapter 10

Network Interface Technology

100Base-TX
The 100Base-X specifications refer to both the copper wire based
100Base-TX and the fiber based 100Base-FX standard. Both of these
standards share a signaling specification called the 4B/5B specification,
which had its roots in the Fiber Distributed Data Interface (FDDI).
With this type of signaling specification, each 4 bits of user data are
then converted to a 5-bit code prior to being placed on the media. This,
obviously, adds overhead to the process and means that there has to be
a signal transmission rate of 125 megabaud to transfer data at the
advertised 100Mbps. This extra bit allows the 5-bit symbols to be laid
out so that there will be periodic transitions of signals to allow the
receiver to maintain the synchronization with the incoming data.
100Base-TX uses two pairs of twisted pair cabling. It uses one pair of
the wires for transmitting data and the other pair for receiving data.
Each end of the cable has an RJ-45 connector on it. If you work with
Ethernet networks, this is the standard Category 5 (Cat 5) cable that
you know and love. This cable is rated to transmit at frequencies up to
100MHz. The 125 megabaud transmission rate that we mentioned in
the last paragraph comes out to a maximum frequency of 62.5MHz, so
there is room left over. This media will also allow for full duplex mode,
if supported. To make use of full duplex mode, the NIC and the hub
must support it and be configured to use it.
All connections on a 100Base-TX network are point to point, with a
transceiver at each end of the cable. Most of these connections simply
link a computers NIC to a hub. If necessary for testing, two computers
can be hooked directly together with a cross over cable. In a cross over
cable, the transmit pair of cables from one system is set to be the receive
pair for the other computer and vice versa. Cross over cables are only
needed when connecting two computers directly together. If you are
connecting the computer with a hub, a straight through cable is fine.
If noise is a problem, or if you are taking over a building that is wired to support
Token Ring, 100Base-TX can also support 150 ohm shielded twisted pair (STP)
cabling. When you are using STP, the RJ-45 connector is replaced with a 9-pin
connector that looks similar to a connector on a serial cable.

100Base-FX
100Base-FX supports 100Mbps rates over two fiber optic cables, instead of
two pairs of copper cable. The maximum segment length for half duplex
links is 412 meters, and if you are using fullduplex links, the range is 2,000
meters. Basically, this is just the fiber counterpart to 100Base-TX. The cable
and connector may be different, but everything else stays the same.
Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Gigabit Ethernet

447

100Base-T4
There is one other type of 100Base-T specification cable. That is 100Base-T4.
This cabling specification can use the older Category 3 UTP cable.
100Base-T4 makes use of all four pairs of cables. One pair transmits
data, another pair receives data, and there are two bidirectional pairs that
can either transmit or receive. Because this form of cabling cannot support
simultaneous transmit and receive functions, it cannot support full duplex.

Gigabit Ethernet

ave you ever heard a network administrator who said he had too much
bandwidth? Me, either. Most administrators are in a serious bandwidth crunch,
and one way of handling that issue is by looking at Gigabit Ethernet. Gigabit
Ethernet is the IEEE 802.3z technology that provides throughput of one billion
bits per second. It is normally used as an enterprise backbone, and it has several
things going for it:


The migration between current implementations can be done in an easy

and straightforward manner with little or no disruption to the network.

There is a relatively low cost of ownership from the perspective of purchase

and support costs.

It has the expansion capability to support both new applications and

new data types.

It offers a lot of design flexibility.

Migration to Gigabit Ethernet

Like the other Ethernet implementations we have discussed in this
chapter, Gigabit follows the 802.3 specifications. That means that it is
like its 10Mbps and the 100Mbps cousins and is backwardly compatible.
Because of the compatibility, your migration to the higher throughput
can be done in a staged manner instead of all at once. All of the speeds use
the same 802.3 frame format, full duplex operation, and flow control
methods. If you operate a Gigabit Ethernet network in half duplex mode,
it uses the same CSMA/CD access methods to resolve contention for the
shared media. Basically, this means that Gigabit Ethernet is just regular
old Ethernet, only a lot faster.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

448

Chapter 10

Network Interface Technology

Switches vs Hubs
Lower speed Ethernet devices can be connected to Gigabit Ethernet devices
using LAN switches or routers. These switches or routers will adjust one line
speed to the other. When we talked about the lower speed Ethernet networks,
we mentioned that computers could be hooked together by running the cables
through a hub.
When you generally think of a hub, you think of a part of a wheel where all
the spokes join together. The same is true with a network. With a LAN, a hub
is where all the cabling comes together and where all the data will pass through
on its journey from source to destination. The design of the hub usually means
that this data will be delivered to every computer that is attached to the hub.
With slower Ethernet technologies, this usually is not a problem. With Gigabit
Ethernet, it would be a serious bottleneck.
A switch does not look much different than a hub. As a matter of fact, at
first glance, the only way to tell which is a switch and which is a hub is to
read the device name or the price tag. They are both collection points for
cables. The difference is that an Ethernet switch is an intelligent device,
meaning that it learns about its environment. Look at Figure 10.9.
FIGURE 10.9

Switching technology

Switch

Host 1

Host 2

Host 3

Host 4

Host 5

Host 6

If we were looking at a hub rather than a switch, if Host 1 wanted to send

data to Host 5, every other host attached to the hub would have to receive and
act on the packet, even if it is just to decide that the message is not for that host
and ignore it. With switching technology, the switch learns the media access control (MAC) addresses of each of the hosts NICs. Now, if Host 1 wants to send
a message to Host 5, the switch looks at the destination address and completes
a circuit that just includes Host 1 and Host 5. The other computers on the

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Gigabit Ethernet

449

network are not bothered with dealing with messages that are not intended for
them. When you consider the virtual circuits that are created every time a packet
is delivered, think of the importance in a technology where collisions can be a
major bottleneck. If collisions are avoided, the bottleneck is eliminated.
Now, dont get me wrong, switches have their drawbacks. First of all, a
switch, as an intelligent device, can cost considerably more than a hub. Secondly,
the fact that the switch has to evaluate and make a decision on each and every
packet it handles creates a delay or latency in the delivery of the packets.

Full Duplex vs Half Duplex

If you look at the specifications for any of the IEEE 802.3 technologies, you
will see the terms full duplex and half duplex thrown around. When two
nodes are connected in a full duplex switched path, they can send and receive
packets at the same time. This is what Gigabit Ethernet uses, as well as
employing other standard flow control methods to avoid congestion and
avoid overloading.
When Gigabit Ethernet operates in half duplex mode, it uses the same
CSMA/CD access methods as the slower forms of Ethernet, but with a
twist. It has enhanced the specification to account for the fast speeds.
Without the enhancement, even the smallest Ethernet packets could
complete the transmission before the station that transmitted the packet
could figure out there had been a collision.

Gigabit Ethernet Physical Cables

These technologies are referred to as the 1000Base-X specifications. We will
look at the fiber based 1000Base-LX and 1000Base-SX as well as the copper
based 1000Base-CX and 1000Base-T standards.
Each of these standards is based on a specification that comes from the
ANSI X3.230-1994 standard for Fibre Channel. These specifications use the
same coding scheme (8B/10B) as Fibre Channel. With 8B/10B, each group of
8 bits of user data is converted into a 10-bit symbol that is transmitted over
the media. The overhead requires that the signal transmission rates be set at
1.25 gigabaud to actually transfer data at 1Gbps. These extra bits do make
up for the extra overhead, because each 8-bit data combination has a unique
symbol assigned to it. Additional symbols are defined to transfer control or
to provide other information. This way, instead of transferring 8 or 10 bits
of data, the system is transferring one symbol.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

450

Chapter 10

Network Interface Technology

1000Base-LX
Here the L stands for long. 1000Base-LX uses long wavelength lasers to
send data over fiber optic cable. These long wavelength lasers operate in
the wavelength range of 1,270 to 1,355 nanometers, and both single mode
and multimode optical fibers are supported. These long wavelength lasers
are more expensive than the short wavelength lasers, but the advantage is
that the signal can cover longer distances.

1000Base-SX
If the L stands for long, it stands to reason that the S stands for short wavelength
lasers. These operate in a wavelength of 770 to 860 nanometers, and support only
multimode optical fiber. They are less expensive than the long wavelength lasers.

1000Base-CX
Each of the previous implementations have been fiber, and with 1000Base-CX we
will start looking at the copper specification. 1000Base-CX uses a specially
shielded and balanced copper jumper cable called twinax or shorthaul copper.
Segment length is restricted to 25 meters, which means 1000Base-CS will connect
equipment in small areas like wiring closets.

1000Base-T
The other copper cable specification was released in June 1999. This standard
supports Gigabit Ethernet over 100 meters of Category 5 balanced copper cable.
It uses the full duplex of the fiber, but it is carried over four pairs of Cat 5 cable.
The rate of 1Gbps is achieved by carrying 250Mbps over each of the four pairs.
Like fiber, symbols are transmitted and received on the same wire pairs at the
same time. Parts of the 100Base-T standards have found their way into the
1000Base-T world. It uses the same type of signaling that is used in 100Base-TX.

Token Ring

hen you think of Token Ring, think of IBM, because IBM invented
Token Ring back in the 1970s, and is still its prime champion. While Ethernet uses
the CSMA/CD method of getting information out onto the wire, Token Ring
places data on the wire using a more refined method called token passing. In token
passing, a 3-bit token is passed from one network to the next in a predetermined
sequence. When each node gets the token, it examines it to find out if there is data
attached, and if the data is intended for that node. If there is no data attached, and
the node has data to send, it tacks the data on the back of the token, and puts the
Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Token Ring

451

whole package back on the ring. If there is data attached, and the data is for the
node, the node strips the data and sends the token back on its way. If the data is
not for the node, the whole package is returned to the wire.
This means that unlike CSMA/CD networks, token passing networks are
deterministic. This means that it is possible to calculate the maximum amount of
time that will elapse before any station may be able to transmit a signal. This
method has several things going for it. For example, each time a node touches the
token, it is completely regenerated and re-sent. This means that the token will not
suffer degradation over the length of the cable.
Take a look at Figure 10.10 to see an example of a Token Ring.
FIGURE 10.10

Token Ring

Workstation

Workstation

Workstation

Token Ring
Workstation

Workstation

Workstation

Workstation
Workstation

Token Ring, while a logical ring, is a Physical Star. All of the computers are
plugged into a Multistation Access Unit (MSAU) using either shielded twisted
pair (STP) or Category 5 unshielded twisted pair cable. The MSAUs can be
linked together to increase the size of the ring. Look closely at Figure 10.11
and see what I mean.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

452

Chapter 10

Network Interface Technology

FIGURE 10.11

Token Ring MSAUs linked together to increase the size of the ring

MSAU
Ring
IN

MSAU
6

Ring
8 Out

Ring
IN

Stations

Ring
8 Out

Ring
8 Out

Stations

Patch
cables

MSAU
Ring
IN

Ring
8 Out

MSAU
Ring
IN

Lobe
cables
Stations

Stations

IEEE 802.5 Specifications

Token Ring is defined by the IEEE 802.5 specifications, which are almost identical to and totally compatible with IBMs implementation of Token Ring.
Basically the IEEE specs have mirrored the development of Token Ring. Take a
look at Table 10.2 to see the differences between the specifications and the
implementations.
TABLE 10.2

IBM Token Ring Implementation Vs IEEE 802.5 Specifications

Category

IBMs Implementation

IEEE 802.5 Specification

Speed

4 or 16 Mbps

4 or 16 Mbps

Nodes per segment

260 on shielded twisted

pair or 72 on unshielded
twisted pair

250

Topology

Star

Not defined

Type of media

Twisted pair

Not defined

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Token Ring

TABLE 10.2

453

IBM Token Ring Implementation Vs IEEE 802.5 Specifications (continued)

Category

IBMs Implementation

IEEE 802.5 Specification

Type of signaling

Baseband

Baseband

Access method

Token passing

Token passing

Encoding

Differential Manchester

Differential Manchester

Networks are formed when Token Ring nodes are directly connected to
the MSAUs and the MSAUs are connected together to form one large ring.
The MSAUs are linked with patch cables while lobe cables connect the
MSAUs to the station. If a station begins to cause problems, the MSAUs can
automatically bypass it by opening a relay.

The original specification called for only one token on any Token Ring. In
some of the newer implementations, there can be two tokens.

Workstation Priority
One of the features that Token Ring has is a sophisticated priority system
that will allow certain high-priority designated workstations to broadcast
more frequently. The Token Ring frame has two fields that work to control
priority, the priority field and the reservation field.
If a workstation gets the token and the workstation has a priority equal
to or greater than the priority value already in the token, the workstation can
reserve the token for the next pass around the network. When the next token
is generated, it will include the higher priority of the reserving workstation.
If a station raises the tokens priority, it must reinstate the previous priority
after their transmission is finished.

Fault Monitoring
One of the workstations on the network is designated the active monitor.
This workstation is the centralized source for timing information for all the
other stations on the ring. The active monitor can be any workstation on
the ring. The active monitor has to do other things besides keep the time.
It must also check to make sure that any frames that are on the ring are not
continuously circling the ring. If it comes across an errant frame, it is up to
the active monitor to discard the frame and generate a new token.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

454

Chapter 10

Network Interface Technology

Another way that Token Ring networks auto detect problems is a technology
called beaconing. Think of it this way. We said that a Token Ring network is
deterministic. That means that workstations know when to expect the token to
come flying past. If it doesnt, apparently there is problem. The workstation will
then send out a beacon frame that lays out all the information the workstation
knows about the failure. This would include things like the station that is reporting
the failure, its nearest active upstream neighbor (NAUN), and everything else in
between. The beacon frame starts a process called auto reconfiguration. When
auto reconfiguration kicks in, the ring is basically reestablishing itself around the
node that is no longer responding.

Token Ring Advantages and Disadvantages

Token Rings tend to be larger than most Ethernet segments, so you can have
more workstations on a contiguous network. When compared to traditional
Ethernet, a 16Mbps Token Ring has a much higher bandwidth and speed.
When compared to traditional Ethernet and Fast Ethernet, the components
that make up a Token Ring network, from the NICs to the MSAUs, tend to be
more expensive.

Network Interface Cards

Some people think the only thing a NIC does is put data out on the network. Not anymore. Depending on the brand and type of network card you buy,
the NIC can put information on the wire and also be an integral part of your network management scheme. There are cards that will allow the computer to boot
without an operating system, to sleep when not being used, and to wake on signal
from a remote server somewhere on the network. In this section we are going to
take a look at some of the technology the cards you may decide to use in your
server can utilize to increase performance and even make your life easier.
All through this book, we have been stressing ways to increase
throughput and increase functionality. When it comes right down to it,
everything else depends on this component to make it look good. You
may have the fastest processor, with the fastest disk subsystem connected
to the highest bandwidth cabling system possible, but if the network card
that makes the connection doesnt meet the same type of specifications,
everything else is wasted.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Network Interface Cards

455

Also as we have mentioned in this book, the most popular form of

bus architecture is the PCI Bus. It is easy to manage and offers many
advantages over the old ISA or EISA or MCA architectures. PCI still has
its bottlenecks, and several network card manufacturers have discovered ways of maximizing the performance of the bus. There are three
areas where bus performance can be improved:


Taking full advantage of the latest PCI chipsets

Using the latest PCI system commands

Consolidating the input/output cycles by adding some logic to the chips

Performance Enhancements
The early PCI Bus architecture had limited the size of data bursts to a maximum
of 64 bytes. While this was plenty of speed at one time, when we moved in the
age of videoconferencing and other CPU intensive business and mission critical
applications, the limitation became a problem. The newer PCI chipsets can let
larger bursts move across the bus to the NIC. These bursts can be up to the size
of a full packet, or up to 1,514 bytes for an Ethernet packet. This packet can be
moved in a single bus mastering operation.
As far as the latest commands go, there are three commands that can
make a difference:


Memory Read Line (MRL)

Memory Read Multiple (MRM)

Memory Write Invalidate

The Memory Write Invalidate command enhances the cache snooping

logic for the Memory Write cycles. This means that the PCI chipset no longer
has to perform certain operations after every transfer, and that allows the
CPU to make more efficient use of its available resources.
Other advances have reduced the number of I/O cycles that the NIC has to
perform by making the card perform more efficiently and performing some of
these operations at the same time, or in parallel. This way the card can clear the
Interrupt Request and reset it in one cycle, rather than requiring several.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

456

Chapter 10

Network Interface Technology

NIC Load Balancing

Even though the bandwidth of the network cabling keeps growing, there
are additional bottlenecks starting to show up. These bottlenecks make it
mandatory for high volume servers to balance their load over multiple
processors. The information, however, still has to get out onto the wire.
NIC load balancing is a technology where multiple interface cards can be
configured to work together as a single connection between the server
and the switch it is plugged into. Now, servers that experience heavy
traffic can distribute the traffic across several links, relieving the network
congestion. Look at Figure 10.12 to see what I mean.
FIGURE 10.12

NIC load balancing

1 Gb/s
Desktop

Enterprise Switch

Switch
100 Mb/s

400 Mb/s
Aggregated Link

Desktop

Layer 3 Switch
1 Gb/s
Server with Four
Load-Balancing
NICs

Desktop

1 Gb/s

Desktop

Enterprise Switch

In this case, there are four server NICs that are configured as a NIC group.
Each NIC helps to eliminate some of the traffic coming from the service link.
These NICs provide a scalable performance increase without major investments
in new technologies. The load balancing works on two fronts: transmission and
reception.
Depending on the implementation, each NIC group can have between
two and eight network cards. Each NIC in the group will obviously have its
own Media Access Control (MAC) address, but the entire group is given just
one IP address, forming a virtual NIC that can become part of a virtual local
area network.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Network Interface Cards

457

Transmit Load Balancing

When the system gets ready to transmit information, the data goes through
an advanced hashing algorithm that will distribute the outgoing traffic
evenly across the NICs that make up the load balancing group. The traffic is
assigned to a specific NIC for transmission based on things like the destination IP address and several other parameters that go into the formula. As
each client computer connects to the network, the hashing algorithm makes
the determination as to which NIC the client is going to be assigned to for the
duration of the session. If a NIC within the group is not currently being used
for a connection, the algorithm will automatically change to make sure the
NIC is utilized. Because the assignments are based on the unique client IP
addresses, the transmit load balancing works even with routers and other
Level 3 devices, like a Level 3 switch.

Receive Load Balancing

In this case, the selection of the NIC is based on a round-robin approach.
Incoming traffic is assigned to the next available NIC on a rotating basis.
If a single client has several connections to the server, the inbound traffic
from each connection is shifted in a similar round-robin fashion to the next
available NIC in the group. This process is done by using a variable or
steering IP address that will allow the software to intercept the normal
handshaking process and rotate the servers MAC address for all the NICs
in the group.

Other Load Balancing Methods

Other load balancing methods include distributing the data packets to different
links in a round-robin scheme. This alternative means that packets that will be
handled by different NICs can reach the client in the wrong order, causing
retransmission.
Another alternative is based on the destination MAC address being used
to assign the NIC. The problem with this is that all transmissions to the
router will be assigned to the same NIC.

Resilient Server Links

Another method of increasing the efficiency of NICs is to implement resilient
server links as a method of fault tolerance and fail over. Like the concept of load
balancing, there are multiple NICs configured in the server. One of these links is
designated as the primary link and the other is the backup. Software will constantly monitor the health of the primary server link, and if there is a problem,
it will switch over to the backup link.
Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

458

Chapter 10

Network Interface Technology

Self-Healing Drivers
The software NIC driver is the piece of the communication puzzle that
allows the Ethernet NIC to encode the data that comes in from the operating
system and put it in a form that can be sent over the network. It is also used
to decode the information that comes in from the NIC to prepare it for the
upper layer protocols. Driver development tended to center around adapting
the driver to work with various operating systems, not to check the drivers
health or to see how it was performing on a day to day basis.
Some manufacturers have come up with the Self-Healing Driver, which is
designed to recognize problems with the NIC or with the software driver. These
problems, hopefully, can be recognized and solved before the problem results in
outage. In many cases, these problems can be corrected automatically.

This has been a brief overview of the new technologies for NICs. For more
information, be sure to read Chapter 11, More on NICs, where we will be
looking at these technologies in more depth.

Real World Scenario

The reason that most computer networks run using the Ethernet specifications
is simple: money. Its cheaper than the alternatives, and it flat out works. It is
easy to install and configure, and it is a mature technology where the bugs
have all been worked out.
Since it has been around since God created Dirt, its strengths and weaknesses have been proven and allowances or solutions have been devised for
the weaknesses. Ethernet backbone speeds are commonly gigabit transmission rates, and even to the desktop, the cost of Fast Ethernet has dropped to
the point where it is difficult to buy just a straight Ethernet 10Mbps concentrator anymore. Everything, it seems, has gone to the 10/100 standard.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Configuring Network and Verifying Network Connectivity

459

Configuring Network and Verifying

Network Connectivity

onfiguring the network on a client or a server and then verifying its

connectivity is a big part of an admins daily work. Sometimes the configuration of client computers is left up to the desktop support personnel, but
in smaller shops the admin may be the chief cook and bottle washer and
will have to configure client computers as well.
If you need to connect clients and servers to more than one NOS, your job
complexity increases exponentially. You may have to install a proprietary
client for each different NOS you want to connect to. Then again, maybe
not. It depends on the NOS, the friendliness of the client component, and
whether there are other workarounds or not. For example, suppose you have
a Novell NetWare network and a Windows 2000 (W2K) network and you
need the W2K servers to talk to the NetWare servers and your users to talk
to both. Youll have to decide what client you want to install on your user
desktops and whether youll set up interconnectivity between the W2K and
NetWare computers (very possible) or leave them as stand-alone entities.
The NetWare client provides connectivity with both W2K and NetWare
servers, as does the Microsoft client for networks. For other network operating
systems such as UNIX or VINES, you may have to resort to other, less familiar
clients, or to technologies such as Network File System (NFS), Samba, and other
interconnectivity techniques.
As a general rule of thumb youll need to know several basic pieces of
information before you can configure a client:


If participating in a Windows NT/2000 network, what is the domain

name this client will be joining?

If participating in a NetWare network, what is the name of the tree

and context (or server name if a bindery installation) that this client
will be connecting to?

If participating in a UNIX or VINES network, what is the name of the

server the client will be connecting to?

What is the username?

What is the password he or she should use?

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

460

Chapter 10

Network Interface Technology

What logon scripts (if any) should the client be running at logon time?

What protocols should be used with this client?

Some client components come with the NOS, others need to be downloaded from a manufacturer Web site. For example, if youre running in a joint
Windows 2000Novell NetWare network and you want to use the NetWare
client, youll likely have to download it from the Novell Web site (or obtain it
from your NetWare installation CDs). The Microsoft client comes with the
various flavors of Microsoft personal operating system software.

Verifying Network Connectivity

After setting a client up with network connectivity, you need to perform
some basic tests to make sure the client can successfully connect to the network. For example, if youre running TCP/IP on the client, you might try
some basic ping tests (starting with pinging the loopback address as
127.0.0.1) to make sure the client can get out on the network. If you cant
ping a host, check your IP configuration information to make sure its OK,
then check the client and NIC settings. Oftentimes the NIC isnt initialized
because its trying to use an IRQ thats already in use, or its set for a speed
that the network cannot support.
With non-IP servers such as IPX-based NetWare servers, you might try
connecting to a volume on the server just to make sure you have connectivity. Youll also want to validate that the permissions or rights that the client
is supposed to have for given resources are correct.

Upgrading Adapters

pgrading adapters such as NICs, SCSI cards, and RAID array controller cards is a fairly easy process. Essentially you have two choices. You
can either choose to upgrade the adapters firmware (also called its microcode) or replace the adapter. Depending on what youre expecting out of the
adapter, your decision could go either way.
For example, perhaps youve got a NIC thats acting up, and the manufacturers technical support department has advised you that simply upgrading the
NICs microcode solves the problem youre seeing. Youd go to their Web site,
download the upgrade, and apply it. You may have to apply it to hundreds or

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Summary

461

even thousands of NICs but, hey, its better than replacing thousands of NICs,
isnt it?
On the other hand, if youve got a SCSI adapter thats hanging back in the
SCSI I days and you want to hook up an Ultra-SCSI device to your server,
then youre looking at a hardware upgrade rather than a firmware upgrade.
With an exotic device such as a RAID array controller card or a multiport
serial adapter, you may be able to add certain components to the adapter,
making it more feature-rich. For example, with a RAID card, you may buy
a battery backup module that you need to install on the card or you may
desire to upgrade its RAM module so that it can hold more data in memory
before flushing to disk. With a multiport serial adapter, you may be able to
buy add-on modules that add more serial ports to the adapter.
All methodologies are common as wood ticks in summer, so be prepared
for any eventuality.

Performing an Upgrade Checklist

Its critical, especially with upgrades to adapters, that you plan out your
upgrade strategy, listing the problem youre trying to solve, the method by
which you intend to solve it, and the end result expected. Prepare an upgrade
checklist that exactly denotes what you intend to dorun it by stakeholders
for approval and go forward with your plans.

Summary

n this chapter, we traced the beginnings of Ethernet and followed its

advances through the Gigabit speeds. We have also looked at how Token
Ring can be used and what advantages it offers. Next, we covered some NIC
basics.
We also talked about configuring network client software and making
sure that you have key information available to youthings like the
domain name, server names the client will connect to, username, password,
and other information. Be sure to check client connectivity after installing
a network client component on a client computer.
Finally, we talked about upgrading adapters, why youd want to upgrade,
and the different kinds of upgrades you can perform. A firmware upgrade simply upgrades the microcode software on the device. Some adapters can have
enhanced functionality added to themthings like battery backup modules

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

462

Chapter 10

Network Interface Technology

for RAID cards. You might have to do a wholesale changeout of a card from
older technology to very new technology in order to support a peripheral
youre planning on adding.
In the next chapter, we are going to look at fail over technology for
Network Interface Cards and how these adapters can effectively be teamed
up, and examine load balancing in more detail.

Exam Essentials
Know the basics of Ethernet Ethernet operates at 10Mbps.
Know the basics of Fast Ethernet Fast Ethernet operates at 100Mbps.
Know the basics of gigabit Ethernet Gigabit Ethernet operates at
1,000Mbps.
Know the basics of Token Ring Token Ring operates at either 4Mbps
or 16Mbps.
Know the difference between a cross over cable and a straight through
cable, and when to use each Straight through cables are used to connect
devices to hubs or concentrators. A cross over cable is use to connect concentrators together, or to connect two PCs together without the benefit of
a concentrator.
Know how to choose the right network card for your system bus
Chose the network card that most closely matches your bus speed and
provides you with the greatest flexibility.
Understand client connectivity Know and understand that client computers (and servers) require network connectivity software to allow them to
communicate out on the network. Realize that the complexity of the client
connectivity software issue is relative to whether you have a disparate NOS
environment or not, what protocols are in use, and what servers the client
will connect to. Some NOS products have built-in client components you
can use, while others require a separate download or client installation.
Be able to upgrade adapters Be familiar with the reasons for upgrading
adapters. You might need to upgrade an adapters firmware to make it
more functional. Or you might need to add an add-on component to the
adapter. Or you might simply need to change the adapter out for a newer
model.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Summary

463

Key Terms
Before you take the exam, be certain you are familiar with the following terms:
100Base-TX
1000Base-CX
1000Base-LX
1000Base-SX
1000Base-T
100Base-FX
100Base-FX
100Base-T4
100Base-TX
100Base-X
10Base Fiber Backbone (10Base-FB)
10Base Fiber Link (10Base-FL)
10Base Fiber Passive (10Base-FP)
10Base2
10Base5
10Base-F
10Base-FL
10Base-T
10Broad36
active monitor
Attachment Unit Interface (AUI)
auto reconfiguration
auto-negotiation
bandwidth
bidirectional
bits per second (bps)

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

464

Chapter 10

Network Interface Technology

British Naval Connector (BNC)

Carrier Sense Multiple Access Collision Detection
clock signal
cross over cable
Ethernet
Fast Ethernet
Fiber Distributed Data Interface (FDDI)
fiber optic inter-repeater link (FOIRL)
flow control
frequencies
full duplex
graphical user interface (GUI)
half duplex
Hertz (Hz)
Institute of Electrical and Electronic Engineers (IEEE)
International Organization for Standardization (ISO)
LAN switches
latency
Local Area Network (LAN)
Manchester encoding
Media Independent Interface (MII) connector
multilevel signaling
Multistation Access Unit (MSAU)
nearest active upstream neighbor (NAUN)
NIC load balancing
page-description language (PDL)
Palo Alto Research Center (PARC)
priority field

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Summary

reservation field
resilient server links
routers
Self-Healing Driver
shielded twisted pair (STP)
signal ground
software NIC driver
synchronous clock encoding technique
Token Ring
unshielded twisted pair (UTP)
what-you-see-is-what-you-get (wysiwyg)

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

465

466

Chapter 10

Network Interface Technology

Review Questions
1. Lucinda is a network administrator for a company that has both Novell

NetWare and Windows 2000 servers. Which network clients could she
install to obtain connectivity for a given client?
A. Microsoft client
B. Novell client
C. NetBIOS client
D. NFS client
E. TCP/IP client
2. Youre working with a two-year-old server that has an on-board SCSI

adapter with a SCSI II external connector. Youre getting ready to buy

an external RAID cabinet for some added disk space and are surprised
to find that the cabinet requires an Ultra-SCSI connection to the
servers SCSI adapter. What is the best option for you to go with when
working on this problem?
A. Buy a cable with both SCSI II and Ultra-SCSI connectors.
B. Replace the RAID cabinets SCSI adapter with a SCSI II adapter.
C. Replace the servers SCSI adapter with an Ultra-SCSI adapter.
D. Update the servers SCSI adapter firmware.
3. Ethernet is assigned what IEEE standard designation?
A. 803.2
B. 802.5
C. 802.2
D. 802.3

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

4. Token Ring is assigned what IEEE standard designation?

A. 803.2
B. 802.5
C. 802.2
D. 802.3
5. Token Ring networks and technology are usually associated with

which company?
A. 3COM
B. Honeywell
C. Conrad Thomas
D. IBM
6. Ethernet uses what form of cable access?
A. Token running
B. Token passing
C. CSMA/CA
D. CSMA/CD
E. CSCD/MA
7. How big is a token?
A. 3 bits
B. 3 bytes
C. 3 nibbles
8. How does Token Ring access a network?
A. Token running
B. Token passing
C. CSMA/CA
D. CSMA/CD
E. CSCD/MA

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

467

468

Chapter 10

Network Interface Technology

9. If a collision happens on an Ethernet network, how does the sending

host decide to resend the packet?

A. It immediately starts the sending process over again.
B. The packet is deemed destroyed and the communication is halted.
C. The NIC waits a predetermined period of time before resending.
D. The NIC sends the packet to another source address.
10. What is the maximum speed of a regular Ethernet network?
A. 10Mbps
B. 100Mbps
C. 1000Mbps
D. 1Gbps
E. 100Gbps
11. Token Ring networks can operate at what speeds?
A. 2Mbps
B. 4Mbps
C. 8Mbps
D. 16Mbps
12. How many tokens were on an original Token Ring network?
A. One
B. Two
C. One per host
D. One per server
13. Why would Gigabit Ethernet over fiber usually be used?
A. Provide rapid communication to the desktop
B. As a link between switches
C. As a link between routers
D. As a backbone

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

469

14. What access method does Fast Ethernet and Gigabit Ethernet use to

get data on the wire?

A. Token running
B. Token passing
C. CSMA/CA
D. CSMA/CD
E. CSCD/MA
15. Which card could use bus mastering?
A. 8-bit Ethernet NIC
B. 16-bit Ethernet NIC
C. 16Mbps Token Ring
D. DPCI Ethernet or Token Ring
16. One of Wendys file and print servers has a very old Future Domain

SCSI I adapter in it and she thinks that by replacing the adapter she
can enhance the throughput of the disks and speed up the computers
operation. What are some concerns that Wendy must keep in mind as
she considers the upgrade?
A. Updated cabling
B. IRQ issues
C. Memory I/O concerns
D. Updating new cards BIOS
E. PCI slots in server
17. Which two cable types can be associated with Token Ring?
A. UTP
B. STP
C. Cat 2
D. Cat 5

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

470

Chapter 10

Network Interface Technology

18. An RJ-45 connector is usually associated with __________.

A. ThinNet
B. Thicknet
C. UTP
D. STP
19. A vampire tap is usually found on what type of implementation?
A. ThinNet
B. Thicknet
C. UTP
D. STP
20. Which cable type would you normally associate with a long distance run?
A. UTP
B. Thicknet
C. STP
D. ThinNet
E. Fiber

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Answers to Review Questions

471

Answers to Review Questions

1. A, B. When working with NetWare and Microsoft networks, Lucinda

could choose to install the Microsoft client and set up connections to

NetWare servers or install the NetWare client and set up connections to
NT/2000 servers. In most installations Ive worked with like this, the
administrators typically opt to install the NetWare client and then allow
for Microsoft connectivity as well.
2. A. The servers SCSI adapter is built into the motherboard so you cant

upgrade it. You could add a SCSI adapter to the server, however. Chances
are that the same is true for the RAID cabinetits SCSI connection is
wired into the backplane of the cabinet. Upgrading the servers SCSI card
firmware wont help with connecting the SCSI II to the Ultra-SCSI, however. Your best and most cost-effective option is probably to buy a cable
thatll handle the two different connection types.
3. D. The IEEE standard for Ethernet is designated as 802.3.
4. B. The IEEE standard for Token Ring is designated as 802.5.
5. D. IBM invented Token Ring, and the IEEE standards mirror the IBM

implementation.
6. D. Ethernet use Carrier Sense Multiple Access Carrier Detection

(CSMA/CD).
7. A. A token is 3 bits.
8. B. Token Ring uses the token passing technology.
9. C. The NIC waits a predetermined period of time before resending

the packet.
10. A. The straight Ethernet specification is for signaling traveling at

10Mbps.
11. B, D. Token Ring network can operate at 4Mbps or 16Mbps.
12. A. In the original specification, there was one token per ring.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

472

Chapter 10

Network Interface Technology

13. D. Gigabit Ethernet over fiber is usually used as a backbone.

14. D. Fast Ethernet and Gigabit Ethernet use CSMA/CD to get data on

the wire, just like regular Ethernet.

15. D. Bus mastering will be done by PCI devices. The other devices are ISA.
16. A, D, E. The new card should ultimately be set to use the old cards

IRQ and memory I/O, so there shouldnt be an issue there. She may
have to have different cabling for the new card. If you buy a new SCSI
adapter, generally you get a cable that allows one or at the most two
devices to be added onto the chain. But Wendy may have several
devices and will have to provide an additional cable. Also shell probably want to consider verifying the cards BIOS version and updating
it if its older than the current one. The most important concern she
should have will be verifying if there are available PCI slots in the
server. The Future Domain adapter is likely to be an ISA or EISA card
but most of todays cards are PCI. This could be a big concern if the
server is an older pre-PCI unit.
17. B, D. Token Ring will use either Cat 5 or shielded twisted pair.
18. C. An RJ-45 connector will usually be on an unshielded twisted pair

cable, regardless of category.

19. B. Vampire taps are used on Thicknet networks.
20. E. Fiber is usually associated with long distance runs.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Chapter

More on NICs

11

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

e touched on some of this material at the end of Chapter 10,

Network Interface Technology, but the subject deserves to be covered in more
detail. The way to handle that is to give it a whole chapter. Having multiple adapters in a server is not a new technology, but using the adapters to Load Balance,
Team, or provide Fault Tolerance is a benefit that every network administrator
should know about.

Although this chapter does not directly cover the objectives listed on CompTIAs
Server+ exam blueprint, it does provide valuable information for your fieldwork.
This chapter complements some of the topics addressed in Chapters 1 and 10.
True to our word, we are concerned with your overall knowledge, not just what
you need to know to pass the exam.

Adapter Fault Tolerance

s I mentioned in the last chapter, no matter how well everything else on

the network was designed, installed, and managed, it would be meaningless if
the network adapter card linking a key server to the network were to fail. Just
like disk subsystems that can be fault tolerant, NICs can enjoy the same benefits.
There are many different methods of providing Fault Tolerance, but
whether you call it Adaptive Fault Tolerance, as CompTIA does on their
Web site, or Adapter Fault Tolerance, the solutions all fall into one of two
camps. They are either hardware solutions or software solutions. The hardware solutions usually include at least two adapters and drivers installed in the
server. Hardware solutions are offered by many of the major computer hardware vendors, like 3Com, Compaq, Intel, Madge, IBM, and Olicom. The software solutions allow you to load the fault tolerant software and drivers to

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Adapter Fault Tolerance

475

work with existing implementations from other vendors to provide redundancy among multiple adapters. As you would imagine, before purchasing any
solution, you should make sure that it works with the make and model of network adapter that you have in place. Some vendors will say that their software
will work with any adapter. If the software solution you choose makes this
claim, it would be my suggestion that you spend extensive time in the lab
checking out their claims.
Adapter Fault Tolerance is operating systemindependent. It is supported
by Windows NT, Windows 2000, Novell Netware, and UNIX because the
technology is provided by the network card vendors. What this technology
has in common is that in each case, the solutions will use the high-speed PCI
network card that works with PCI Bus Mastering or a PCI Bridge. In turn,
each implementation will have minimum hardware requirements for the
servers CPU power and memory requirements.

So far, we have been talking about a server that uses multiple network cards as
a way to provide fail over support. There are some manufacturers that have taken
a different approach. At least one manufacturer has created a network card that
has what amounts to four network cards consolidated into one PCI device. This
device provides the server administrator with a way to maximize the number of
PCI slots in the server. While this is a way to get the most bang for your PCI slot
buck, it does not solve the single point of failure question. If the individual card
fails, instead of having one network segment out of commission, you have four.

Infrastructure Concerns
As you plan your Fault Tolerance solutions, keep in mind that the solution
extends beyond the server. For every fault tolerant network card added to the
server, there must be an equal number of concentrator or switch ports to provide
the actual connectivity to the network. You must also ensure that the speed of
the concentrator or switch matches the speed of the fault tolerant adapters.
There are differences between concentrators and switches. The largest
difference is in communication methods. Look at Figure 11.1.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

476

Chapter 11

More on NICs

FIGURE 11.1

Network using a concentrator

Workstation

Workstation

Workstation

Concentrator

Server

Workstation

Workstation

As you can see, all the workstations are plugged into the concentrator,
just as they would be plugged into a switch (a concentrator can also be
referred to as a hub). The concentrator concentrates the signal, so that all the
computers plugged into it will receive the same packets. This means added
processor overhead at each workstation. The processor must determine if the
packet is intended for that workstation. If it is, the workstation must act on
the packet. If not, the packet must be discarded. This also means more network traffic. Instead of sending a signal to just one host, the server is sending
a signal that will go to dozens of hosts.
A switch, on the other hand, is an intelligent device. It creates a virtual circuit
with each workstation. If the server wants to send a packet to a workstation, the
switch will create a virtual circuit so that only that workstation sees the packet.
That reduces the overhead for all the other workstations. This is a two-way
street. If the host wants to send a packet to the server, the switch will create a
virtual circuit back to the server.

Switches are not usually used in a server/workstation communication process. A

switch would be used to connect a server to a group of smaller concentrators to
more effectively manage the network infrastructure bandwidth.

Here are some other things to keep in mind:

Concentrators only work in half duplex mode, while switches will

work in either half duplex or full duplex mode. While all adapters will
support half duplex mode, if the adapter is going to be paired with a
switch, you will want to make sure that the adapter and the switch are
configured to run in full duplex mode.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Adapter Fault Tolerance

477

Speed is another consideration. If your concentrator or switch will operate at 100Mbps, you may want to make sure that auto-negotiation is
turned off at the adapter. Auto-negotiation is the handshake between
the concentrator and the network card that determines the speed the
card will operate at. While auto-negotiation usually works as advertised, it can force the card to come in at a slower speed (10Mbps rather
than 100Mbps), thus slowing things down.

If you are planning on using fault tolerant adapters, you may enhance
the servers uptime by making sure the adapters and the server support
hot-swapping PCI adapters. In order to replace a card, the slot must be
powered off, the card replaced, and the slot powered back on again.
The replacement adapter should be the exact same make and model as
the failed adapter. This ensures that the drivers that are being used will
be the same, and that ensures that the server does not have to be
rebooted just to load a new NIC driver.

Fault Tolerance Tips and Tricks

When you look at the Fault Tolerance capabilities of an adapter, there are
other issues that need to be addressed. Some of these are:
Transmission Protocols Make sure that the Transmission Protocols
solution you choose supports the protocols you use on your network.
Most hardware solutions are protocol independent, but the software
solutions may not match your protocol needs. The solution should
support IP, IPX, and NetBEUI at minimum.
IP Service Protocols What is the server going to be used for? Some services,
like using the server as a Bootstrap Protocol (BOOTP) server or a Dynamic
Host Configuration Protocol (DHCP) server, may or may not be supported
by a software implementation of Fault Tolerance. Since DHCP is a newer,
more widely used protocol, support will be easier to find than with BOOTP.
Multiple Subnets If more than one subnet will be supported by the server,
make sure that the solution you choose supports network segmentation.
Some solutions have network adapters configured so they serve different subnets. Others may require that all adapters serve the same subnet.
Failure Alerts While having fail over support is a very positive thing, the
system administrator still needs a Failure Alert to know when an adapter
fails. If an adapter fails, the server is running with only one active NIC, a single point of failure, and no redundancy. When you start looking at Adapter

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

478

Chapter 11

More on NICs

Fault Tolerance, find out the kinds of alert mechanism employed and make
sure it fits your lifestyle. If I were doing it, I would make sure that I could
receive an e-mail or a page. Some implementations will only write an entry in
the error log, and that just wouldnt work for me, because I diligently read
error logsfor about a day. Then they are just spot-checked at best. This may
require activation of another TCP/IP protocol, Simple Network Management
Protocol (SNMP). Be sure to test the notification procedure to make sure it
works as advertised by disabling one of the adapters. You should make sure
the notification process works as designed during a controlled outage, not
find out it doesnt work during an unexpected outage.

Scalability
Having the ability to have one of your NICs fail over to a fault tolerant
backup is a great thing, but it also means that two of those precious PCI slots
have been used. Besides the physical hardware constraints, there are other
limitations on expansion that should be considered.
Limited PCI Slots Deciding on your Fault Tolerance levels is a great
thing to do before you decide on server specifications. If you opt to have
two network cards as a fault tolerant pair in the server, that may have a
direct impact on the installation of SCSI cards or RAID controllers.
Vendor or Operating System Limitations Some server operating systems
or software vendors may have limitations on the number of adapters or ports
that can be configured for Fault Tolerance. In some cases, the solution may support only 2 ports. Others may support up to 16 different ports. In addition
there are some vendors that will let you add additional adapters, as long as you
do it two at a time. Other vendors may specify that you can add only an odd
number of adapters at a time, while finally, there are some that require you to
install exactly the same brand, make, and model of adapters in the server.
Memory and CPU If you remember back to the discussion on memory,
you will remember that server memory is used to store data that is
requested by either the operating system or by users. If you add additional
active links to the server, the demands on both the memory and on the
CPU will increase. If these demands increase too greatly, the CPU and the
memory may become a bottleneck.

Cost
When I worked for a value added reseller, I would get really frustrated at times
with the purchasing department. I would do a lot of research on the products that

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Adapter Fault Tolerance

479

the client needed, and I would make a decision based on a business case scenario.
In other words, the client wants to accomplish this. In order to do that, the best
hardware solution is A. I would then fill out the purchase order and submit the
signed and approved purchase order, only to receive a completely different set of
parts. When I asked, invariably I would get the response that the part I received
was cheaper than the part I wanted. The descriptions were similar, and therefore
cheaper was better. Obviously, this is not always the case.
This is especially true when you start to look at Fault Tolerant Adapters. The
suggested retail price should not be the deciding factor. We mentioned earlier
that there are network cards that have four ports on them. If I have a hardware
solution that will let me set up two four-port cards in a fault tolerant configuration, this is going to be more expensive than two one-port cards. It may even be
more expensive than eight single-port cards. While the other solutions may be
cheaper, they do not solve the business problem of providing at least four fault
tolerant links to the network.
When you decide to opt for Fault Tolerance, you should do your homework
and find out what the cost per port is for various solutions. Be sure to include
both the hardware and software implementations in your chart. The chart may
look something like Table 11.1.
TABLE 11.1

Cost Estimates of Hardware-Based Fault Tolerant Solutions

Number of
Ports on
Adapter

Price Per
Adapter
Low

Price Per
Adapter
High

Price Per
PortLow

Price Per
PortHigh

$75

$200

$75

$200

$250

$350

$125

$175

$540

$800

$135

$200

Now, all these numbers are just guestimates of the actual cost. You
would have to fill out an actual chart using actual numbers, and then you
may be able to get a feel for the price breakdown. You can also add in
what a software solution would cost to give you a better idea of all your
alternatives.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

480

Chapter 11

More on NICs

Other Costs
Various companies call this various things, but total cost of ownership
(TCO) certainly fills the bill here. The purchase price of the network cards
or software may be a small part of the actual cost of ownership. Maybe you
will need to have additional wiring run for the redundant link. There is the
time involved in researching, purchasing, installing, testing, and maintaining
the new equipment. Because this is a redundant link, there will have to be an
additional shared media hub, or a separate switch or concentrator port,
which means there are fewer ports available for workstations.
There may even be some hidden costs, like the impact on the CPU. If
the impact on the CPU is too high, you may be forced to either upgrade
or replace the server. Not paying attention to any or all of these things
may mean that you have a fault tolerant NIC solution that is bringing
down performance in other areas.
Performance will bring us to another area. If we are going to have two network cards servicing the same network segment, we should be able to get something out of it besides Fault Tolerance, shouldnt we? Maybe we can have both
the cards service the segment and balance the load between the cards.

Adapter Load Balancing

What we are trying to do with Adapter Load Balancing is to have multiple network cards in the server that handle approximately the same amount of
network traffic. By ensuring that one card is not overworked while another sits
idle, we can improve the overall performance of the network.
This Load Balancing can be accomplished in several ways:

Splitting the network segment. This Load Balancing method

requires adding a network card to the server, plugging the new card
into a different concentrator, and moving workstations from one
network to another. This process is relatively simple and painless if
you are using Novells IPX transport protocol. If you are using
TCP/IP, additional steps are neededbut we are getting ahead of
ourselves. We will talk more about that later.

Using hardware based solutions like a switch.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Adapter Load Balancing

481

Using a NIC that is specially designed for a server and has the capability
of running in a Load Balanced mode.

Using a third-party software solution.

When you start talking about Load Balancing, you need to be careful to define
terms. In this case, we are talking about Network Adapter Load Balancing. If
you start talking Network Load Balancing with Microsoft people, you are now
talking about clustering servers. Different goals and different solutions.

Splitting the Network Segment

How much do you know about Ethernet? Earlier in the book, I talked a
little about the way that Ethernet accesses the network media, using a
method called Carrier Sense Multiple Access/Collision Detection (CSMA/
CD). If all the workstations are plugged into a single concentrator,
CSMA/CD works very efficiently as long as the number of workstations
remains relatively small, like under 50. Once you start getting over 50
workstations on a busy network, the traffic grows to the point where
collisions become more frequent. As the number of collisions grows, the
performance starts to degrade, and as performance degrades, people start
to complain.
Look at Figure 11.2. In this graphic, we have a network segment with 100
workstations on it.
FIGURE 11.2

Single network segment with 100 workstations

Workstation

Workstation

Workstation

Workstation

Workstation

Workstation

Server

Workstation

Workstation

Workstation

Workstation

Workstation

You, as the administrator of this network segment, have become sick and
tired of people complaining about the speed of the network, so you have decided

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

482

Chapter 11

More on NICs

to do something about it. You went to your bosses and proposed breaking the
network up into several segments using Fault Tolerant Load Balancing network
cards, new switches, and new concentrators. Total cost of the solution is
$10,000 (please be advised the figure quoted here has no basis in reality, and is
just a WAG). (For those of you in the audience who are unaware of the acronym
WAG, it is a highly technical term meaning wild a** guess.) When your bosses
review the solution, they applaud the effort to improve service to the network
users, but somehow just cannot find the $10,000 in the budget to make this solution work. However, they have put their heads together and decided that you
should do this, but you should do it for under $750. Bosses do this kind of stuff!
All the time. It makes you crazy.
Anyway, back to the drawing board. You decide that since this is going to
be done on the cheap, what you will do is add a second and third network
card to the server (cost: less than $150), add two concentrators (cost: $550),
and use the other $50 for a well deserved pizza and beer party for the IT staff.
Figure 11.3 is what you come up with. Just assume that there are about 33
workstations on each segment.
FIGURE 11.3

Divided network

Workstation

Workstation

Workstation

Workstation

Concentrator

Workstation

Workstation

Workstation

Workstation

Concentrator
Server with 3 Cards
IBM Compatible

Concentrator

Workstation

Workstation

Workstation

Workstation

Now, granted, this is not a very scientific solution to the problem, but it
does remove the load from one network card and spread it out over multiple
network cards. What is the downside?

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Adapter Load Balancing

483

Well, first of all, this solution is really easy to implement if the only
transport protocols you use on your network are Internet Packet Exchange
(IPX) and Sequenced Packet Exchange (SPX). In the case of IPX/SPX all
network addressing comes from the network card, and the node address is
the Media Access Control (MAC) address of the network card, so you can
reconfigure network segments on the fly with very little problem. If, however, your network, like most networks, uses TCP/IP, the solution becomes
somewhat more challenging.
With TCP/IP, you have just created two additional sub-networks. That
means your IP addressing scheme may have to be reworked or, at the very least,
you will have to set up a DHCP server to service the new network segments. This
is not an insurmountable task by any means; as a matter of fact it is really pretty
simple. The tricky part comes in when 66 (at least) of those 100 workstations
have to be restarted to acquire the new addressing information.
If you are planning on using this solution, plan carefully. Look at what is on
the network and make some informed decisions about what will and what will
not be moved between network segments. For example, if you are hosting your
companys Web page, it makes sense to keep the Web server on the same network segment and not move it. Re-addressing a Web server means changing
DNS entries and almost certainly means some time when the information is
not available. The same is true with e-mail servers, application servers, DNS
servers, FTP servers, or anything else that requires users be able to find that
server at a particular address.
The advantage of this solution is usually speed. Because there are no special
drivers, switches, or software involved, the performance does not hit a snag
while some piece of the communication puzzle decides which network card the
packet should be sent from. The disadvantage of this solution is that it is more
difficult to configure and there is no Fault Tolerance. You still have one network
card managing a network segment and this causes a single point of failure.
While this is one solution, it is certainly not the only solution or even the
best solution. One other solution is a switch that can handle Load Balancing.

Switches, Routers, and Load Balancing

Using switches for Load Balancing is not usually done at the NIC level. This
technology is usually reserved for Load Balancing communications between
high-demand servers, like when you are designing a network to host a very
busy intranet.
The difference between the Load Balancing Ethernet switch and a standard
Ethernet switch is simply one of functionality. A switch will have several ports

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

484

Chapter 11

More on NICs

and usually have lots of processing power, so the switch can forward lots of
packets very quickly. This makes switches one of the solutions to consider for
high-traffic sites or intranets. In these scenarios traffic is going to be arriving
from a high-speed LAN connection, rather than coming in over a slower connection from the Internet
At this level of the hardware infrastructure, you could also install a Load
Balancing router. The difference between the switch and the router is in the
details of what each will support. Switches are pretty easy to figure out, while
routers on the other hand can be a little more difficult to get your hands
around. First of all, what is the router? In this example, it can be a softwareonly package that sells for a couple of hundred dollars or a preconfigured
hardware device that carries a pretty hefty price tag.
Load Balancing routers are generally PC-based, running on either UNIX or
Windows NT/2000. The router will have at least two network cards. One of the
cards will be used for incoming traffic and one of the cards will distribute that
traffic to the servers. The standalone router that is not PC-based doesnt usually
look like an Ethernet switch; instead of having a multitude of ports, it usually
has just two, and it runs more-sophisticated software. Running more complex
software means that a router will have more processing power and more RAM
than a switch will have. If this is the case, obviously the price will go up. Routers
are more suited to those sites where the users will need to be connected to the
server for long periods of time, like hours instead of the usual minutes. Since
these routers also come with proprietary operating systems, if you keep the
patches up-to-date, the router solution should be less vulnerable to hackers.
But what do they do? Look at Figure 11.4.
Lets assume that all the servers behind the switch or the router are Web
servers servicing your Web site. All Web sites will have just one IP address,
but multiple servers may be necessary to accommodate all the hits the
server takes. So, the Load Balancing device takes a look at the incoming
packets and then forwards the packet to any of the servers that can fulfill
the HTTP request. The way they determine which device to send the packet
to depends on a variety of factors that are plugged into an algorithm. This
may be a very simple decision, such as a round-robin, where Server 1 is
first, Server 2 is next, Server 3 comes after Server 2, and so on. It may also
be a method of determining how many active connections a particular
server has open, or even a method of determining how long it will take to
service a particular request.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Adapter Load Balancing

FIGURE 11.4

485

Using switches or routers for Load Balancing

Server

Internet

Server
RI

RO

Load-Balancing
Router or Switch

Server

Server

All requests are not the same. For example, if your servers are going to be
used to fill out forms or for SSL transactions, these present some special
problems. The user will need to get to the same server for the entire session
so the server can keep track of where in the process the user is.

One of the reasons for using switches or routers is to improve response

time. There is also a built-in redundancy factor here. If one of the servers
should fail, the requests will then be sent out among the remaining servers.
Each of the switches supports Packet Internet Grouper (PING) and Internet
Control Message Protocol (ICMP) that will check to see if the server
responds to a request. Some switches will also be able to handle geographical
Load Balancing. You can set up several clusters of Web servers that are
spread all over the world, and you can configure switches or routers to check
to see which servers are busiest and also which servers are closest to the user.
While this solution may be ideal for networks that support large Web
server farms, we also need to look at the basic server level. Lets look next
at server NICs.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

486

Chapter 11

More on NICs

Server NICs
There are network interface cards that are designed to work in a server. These
cards bring more features to your server, although the cost is higher than the
$19.95 white box generic NIC that you can buy at your local computer retailer.
Some of these server NICs are relatively inexpensive (under $175) and come
with the technology and software drivers necessary to support Load Balancing.
In this type of technology, you simply add multiple server-optimized NICs
to your server and then configure the NICs as a group. Each NIC will handle
its share of packets, which will eliminate congestion at the server link. This
means that you should have an incremental and measurable performance
improvement, without having to redesign your entire network.
As I mentioned in the last chapter, the manufacturer defines Load Balancing at the NIC level. Load Balancing methods, for example, can use
transmit and receive Load Balancing to manage the information traffic
inbound and outbound. NICs can be added in clusters of from two to
eight NICs, depending on the slot availability. Each NIC continues to
communicate using its own MAC address, but the entire group is lumped
together using a single IP address, creating a virtual NIC that can be part
of a virtual LAN.
With transmit Load Balancing, the algorithm sends outgoing traffic
evenly across the NICs in the Load Balancing group. The destination IP
address and several other parameters are used to decide what traffic is sent
to which NIC. When a client connects, the algorithm determines which NIC
the client will communicate with and that link will last for the duration of the
connection.
If one of the NICs is not being used, the algorithm will change to make sure
that the NIC will be utilized. Because these assignments are based on the
unique client IP address, this method of Load Balancing will work when traffic
crosses routers or switches.
When traffic is received, the round-robin method of distribution can be used
to disseminate the packets. Each packet that comes in is automatically assigned
to the next available NIC on a simple rotating basis. If a single client has established several connections to the server, the inbound traffic from each connection will shift in this round-robin fashion to the next NIC in the group. Packet
dissemination is accomplished by using a variable or steering IP address that will
allow the software drivers to intercept the handshake process and rotate the
servers MAC address between all the NICs in the group.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Adapter Teaming

487

Another Load Balancing method distributes the data packets by using the
destination MAC address to select a link. The problem with this method is
there is nothing to make sure that the server traffic is balanced across all the
NICs in the group. Also, if multiple clients are connecting from a router, the
routers MAC address will be the one used for allocation. That means that
a single NIC will be assigned to handle the requests from an entire subnet.
The grouping is usually handled through the setup for a NIC or with a special driver that is supplied with the network card. The last solution we are
going to look at involves third-party software products.

Software Solutions
If NIC card manufacturers provide the software that will allow the grouping
of NICs for Load Balancing, why would you even think about paying extra
money for a third-party solution? Well, when you group NICs together,
workstation communication may still tend to communicate toward the NIC
with the lowest MAC address. That means that one of the cards gets to do
the majority of the work, while the other cards remain underutilized.
The software products work by using a specific protocol driver that runs
between the TCP/IP protocol stack and the network cards driver. The administrator can choose which network cards make up the cluster. The software
drivers vary in size and options. The actual Load Balancing process can also
vary, some products equalizing the number of connections between cards, and
other products actually monitoring the amount of traffic to equalize the
uneven workloads over a period of time.
The software solutions also have the capability of ensuring Fault Tolerance.
If one of the NICs goes down, the communications from that card are redirected
to another card in the group.
Finally, these products allow you to monitor the cards performance
with SNMP. The variety of functions that can be monitored depends on the
implementation.
Cost of the third-party products varies. It is usually priced on a
per-server basis.

Adapter Teaming

o far we have looked at the ways that adapters can be configured to

provide fail over protection and a way for adapters to Load Balance. In this

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

488

Chapter 11

More on NICs

section, we are going to look at ways for adapters to combine some of those
characteristics and work together to increase the amount of throughput that
you can provide back to the network. This is referred to as Adapter Teaming.
Depending on the types of adapters you are using, the technology can automatically balance the full duplex traffic among multiple adapters without
any client or network reconfiguration. Since the incoming and outgoing
server traffic is balanced, it can be scaled in increments of 200Mbps, with
totals up to 800Mbps in Fast Ethernet environments and up to 8Gbps in
Gigabit Ethernet.
Technology like this is not used on your ordinary run of the mill print
server. This technology is for the high-performance servers that are running
demanding applications such as these:

Enterprise servers

Web servers

Intranet servers

High-end graphics imaging and rendering servers

Not providing enough bandwidth from the server to the communication

channel can lead to a server bottleneck.

Preventing Server Bottlenecks

We have already looked at one way of dealing with server bottlenecks, by
just adding an additional NIC and re-segmenting the network. This does
reduce the traffic volume on each link and eliminates the bottleneck, but
it also causes those problems we discussed earlier. You have to reassign
network addresses and provide additional hardware. There is also the
loss of Fault Tolerance. If one of the network cards fails, you will have an
entire network segment down.
We also looked at Load Balancing. That works to prevent the server bottleneck by reallocating network traffic and making sure that all the adapters
are carrying their fair share of the load. Since most of the communication on
a network comes from the server to the workstation, this helps to ease the
server bottleneck. Load balancing also provides for fail over support.
Adapter Teaming is like Load Balancing, only it usually requires switch
support. With Adapter Teaming, full duplex transmission is enabled on up
to eight network adapters. There is no network reconfiguration necessary.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Ethernet Adapter Best Practices

489

Unlike Load Balancing, this is done as long as there is a switch in place that
supports the technology.
Look at Figure 11.5 and see how this works.
FIGURE 11.5

Adapter Teaming

Up to Eight
Server Adapters

Up to 800 MBP/s at full-duplex (Fast Ethernet adapters)

Up to 8 Gbp/s at full-duplex (Gigabit adapters)
Hubs
All channels
transmit and receive

Switch Must Support

Adapter Teaming

As you can see, by linking or Teaming the adapters together with the
switch, you can effectively create massive amounts of bandwidth, up to
800Mbps using Fast Ethernet cards. Of the combinations discussed in this
section, this is the most expensive, due to the combination of the specially
designed Ethernet cards and the switch.

Ethernet Adapter Best Practices

In this section, we are going to look at some of the ramifications of

using the technologies mentioned in this chapter:

These technologies are not necessarily independent. For example,

with some manufacturers, a single driver will provide the software
support for Fault Tolerance, Load Balancing, and Teaming. The
agent configuration in the environment will determine which of the

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

490

Chapter 11

More on NICs

features is enabled. In this case, if Load Balancing or Teaming is

enabled, Fault Tolerance is automatically enabled.

High-end adapters must support high-end operating systems. You

will usually find drivers and software available for Novell NetWare
and Microsoft Windows NT/2000. UNIX support will depend on
the version of the operating system.

How alerts are handled will depend on the operating system. For example, in a Windows server environment, alerts will be written to the event
logs. For Novell NetWare, a NetWare alert will be generated. An SNMP
management agent can detect the alerts and trigger a notification via
e-mail, fax, pager, or cell phone.

Real World Scenario

This finishes the chain! Throughout the book you have seen ways that you
can build in redundancy to all the components that make up your server. The
same is true with network cards. You have always been able to add multiple
network cards to a server, but with these high availability network cards, you
can configure them so they will work together to provide fault tolerance, balance the load coming off a busy network segment, or team up to provide
greater throughput.
Why would you use each of these technologies? Well, fault tolerance is a
given. Since it is a part of the other two anyway, it should show how
important it is. No longer is the network card the weak link in the communications chain! Network cards tend to have a higher failure than other
server components, so now, if one fails, communication does not stop.
I would serious consider using load balancing in an application or database server where there is a history of exceptionally high utilization. This
will help manage the communication to the processor and not cause an
entire section of the network to be overlooked.
I think NIC teaming is a perfect solution if you are using a storage area network (SAN). Most of these SANs are IP addressable, and ideal for increasing the throughput to your data storage devices.
Again, each piece has its place in the business case for your server.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Summary

491

Summary

In a high-end environment, you need high-end solutions to provide for

maximum availability. This chapter rounded out the configuration of an
ideal server. After reading this book, you should be able to make informed
decisions on the technologies that you put into your server, from the type
and number of processors, to the amount of memory, RAID configuration,
and network cards. You should be able to decide what services you want
your new server to perform and the appropriate operating system. All in all,
if you studied hard, you should be ready to pass the CompTIA Server+ exam,
and my work is completed!

Exam Essentials
Know how fault tolerance can be instituted in network adapters Fault
tolerant network adapters can be used so if one NIC fails, another will
take its place.
Know how load balancing can be instituted in network adapters Network
adapters for highly utilized servers can be load balanced to ease congestion.
Know how adapters can be teamed Network adapters can be teamed to
provide better throughput

Key Terms
Before you take the exam, be certain you are familiar with the following terms:
Adapter Load Balancing
Adapter Teaming
algorithm
auto-negotiation
Bootstrap Protocol (BOOTP)
Carrier Sense Multiple Access/Collision Detection (CSMA/CD)
concentrator
Dynamic Host Configuration Protocol (DHCP)
Ethernet

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

492

Chapter 11

More on NICs

Failure Alert
Fault Tolerance
hot-swapping
Internet Control Message Protocol (ICMP)
Internet Packet Exchange (IPX)
Load Balancing
Media Access Control (MAC)
network segmentation
Packet Internet Grouper (PING)
round-robin
Sequenced Packet Exchange (SPX)
Simple Network Management Protocol (SNMP)
steering IP address
switch ports
total cost of ownership (TCO)
Transmission Protocols
transmit and receive Load Balancing
virtual circuit
virtual LAN
virtual NIC

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

493

Review Questions
1. To institute any of the technologies in this chapter, you must have a

minimum of how many network cards in your server?

A. One
B. Two
C. Three
D. Four
2. You have three network cards in your server. One of the network cards

fails. The failure is detected almost immediately, and all communication

is transferred over to one of the two remaining network cards. This is an
example of which type of technology?
A. Adapter Load Balancing
B. Adapter Teaming
C. Adapter Fault Tolerance
3. Which of the three adapter technologies may require a special switch?
A. Adapter Load Balancing
B. Adapter Teaming
C. Adapter Fault Tolerance
4. You have five adapters in your server. When packets come in, they are

routed to the network cards in the following order: NIC_1, NIC_2,

NIC_3, NIC_4, NIC_5, NIC_1, . This is an example of what type of
Load Balancing?
A. Random
B. Least resistance
C. Rapid convergence
D. Round-robin

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

494

Chapter 11

More on NICs

5. You have designed a Windows NT 4 server to have fault tolerant network

cards. In addition, your server supports Hot Plug PCI technology. You
have configured your server to use SNMP and to page you when a network card fails. You receive the page and replace the Intel NIC with a
NIC manufactured by 3Com while the server remains available. The new
network card, however, will not send and receive packets. This procedure
was doomed from the very beginning. Why?
A. The card is not seated in the slot correctly.
B. The slot was not powered back on when the card was added.
C. The cable is not correctly installed.
D. The network card driver is not loaded.
E. All of the above.
6. What is the maximum number of network cards that can be Teamed,

as discussed in this chapter?

A. 12
B. 8
C. 4
D. 2
7. You have configured four network cards to operate as a team. What

have you also instituted?

A. Adapter Load Balancing
B. Adapter Teaming
C. Adapter Fault Tolerance

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

495

8. You have a Windows 2000 Advanced Server installation with Load Bal-

ancing implemented across four NICs. By using an SNMP Management

agent, you are able to graph out the number of packets handled by each
of the packets, and you notice that one of the NICs is handling considerably more packets than the other cards. What might be the reason?
A. The adapter that is overworked may have the lowest MAC address.
B. The adapter that is overworked may be the adapter that has picked

up for an adapter that has failed.

C. You have misconfigured your SNMP Management program.
D. There must be a new version of the driver available on the companys

Web site.
9. You have just reconfigured your server, adding three network cards to the

server. You then plug the network cards into new concentrators and
move the patch cables for some of your hosts from the concentrator serviced by the original NIC to the concentrators serviced by the new NICs.
You call users on each of the network segments and they did not see any
interruption in service. Which transport protocol are you using?
A. TCP
B. IP
C. IPX
D. RIP
E. SAP
10. You have eight Gigabit Ethernet cards in your server. These cards are

connected to a special switch, and the estimated bandwidth of the

network is 8Gbps. This is an example of what?
A. Adapter Load Balancing
B. Adapter Teaming
C. Adapter Fault Tolerance

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

496

Chapter 11

More on NICs

11. Ethernet accesses the physical media using what method?

A. CSMA/CA
B. Token passing
C. CSCA/MA
D. CSMA/CD
12. You must use a third-party software solution to institute Adapter

Teaming?
A. True
B. False
13. In which case would Adapter Teaming be inappropriate?
A. Enterprise servers
B. Web servers
C. Intranet servers
D. High-end graphics imaging and rendering servers
E. File and print server for a network of 25 users
14. You have an implementation where the network cards at the server are

not carrying out Load Balancing but it is being managed by a switch

or a router. You are not sure whether the device is a switch or a router.
When you look at it, you notice that it has eight ports. This tells you
the device is which of the following?
A. Router
B. Switch
C. Neither; you still cannot tell
15. You are using a switch between the server and the host. When the server

communicates with the host, the switch creates which of the following?
A. Virtual NIC
B. Virtual Network
C. Virtual Circuit

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

497

16. You have eight NICs in the server teamed together. Your network is

configured to use only TCP/IP. How many IP addresses will the eight
NICs require?
A. 8
B. 4
C. 2
D. 1
17. You have 8 NICs in the server teamed together. Your network is

configured to use only TCP/IP, and you have assigned this group of
NICs one IP address. What have you just created?
A. Virtual NIC
B. Virtual Network
C. Virtual Circuit
18. You have a network card that supports auto-negotiation, but you find

that when you plug the card into a 10/100Mbps concentrator, the card
continues to connect at 10Mbps, even though it is capable of connecting
at 100Mbps. What should you do?
A. Replace the card
B. Turn auto-negotiation off and configure the card to connect at

only 10Mbps
C. Turn auto-negotiation off and configure the card to connect at

only 100Mbps
D. Replace the server
19. A network card can send and receive data at the same time. It is

operating in which mode?

A. Full duplex
B. Half duplex
C. Promiscuous
D. Celibate

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

498

Chapter 11

More on NICs

20. An Ethernet card can only operate in which mode?

A. Full duplex
B. Half duplex
C. Quarter duplex
D. Celibate

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Answers to Review Questions

499

Answers to Review Questions

1. B. You must have at least two network cards in your server.
2. C. This is an example of Adapter Fault Tolerance.
3. B. In some implementations, Adapter Teaming will require a special

switch.
4. D. This is an example of round-robin allocation.
5. D. If you mix manufacturers or if you mix card models, the old drivers

will have to be unloaded and new drivers loaded. With Windows NT

4, this will require the server to be restarted.
6. B. When this book came out, the maximum number of network

cards that could be Teamed was 8.

7. C. Whenever you institute Load Balancing or Teaming, you are also

instituting Fault Tolerance.

8. A. If your network is using TCP/IP, the hosts may gravitate commu-

nications toward the NIC with the lowest MAC address.

9. C. If you were using the IPX protocol, you would not need to recon-

figure the hosts on each of the network segments.

10. B. This is an example of Adapter Teaming.
11. D. Ethernet uses the Carrier Sense Multiple Access/Collision Detection

method of access.
12. B. A third-party software solution may be used for Load Balancing,

but not for Adapter Teaming.

13. E. Adapter Teaming is a solution for a high-demand server imple-

mentation.
14. B. The device is a switch. A router will usually have two ports.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

500

Chapter 11

More on NICs

15. C. The link between a server and a host through a switch will be a virtual

circuit.
16. D. The NICs will require just one IP address.
17. A. You have just created a virtual NIC.
18. C. Turn auto-negotiation off and configure the card to connect at only

100Mbps.
19. A. It is operating in full duplex mode.
20. B. An Ethernet card operates at 10Mbps, and can only operate in half

duplex mode.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Chapter

12

Backup, Restoration,
and Disaster Recovery
SERVER+ EXAM OBJECTIVES COVERED IN
THIS CHAPTER:

3.1 Perform full backup.


Verify backup.

4.1 Perform regular backup.

5.1 Recognize and report on physical security issues.


Limit access to server room and backup tapes.

Ensure physical locks exist on doors.

Establish anti-theft devices for hardware (lock server racks).

7.1 Plan for disaster recovery.



Plan for redundancy (e.g., hard drives, power supplies, fans,

NICs, processors, UPS).

Use the technique of hot swap, warm swap, and hot spare to
ensure availability.

Use the concepts of fault tolerance/fault recovery to create a

disaster recovery plan.

Develop disaster recovery plan.

Identify types of backup hardware.

Identify types of backup and restoration schemes.

Confirm and use off site storage for backup.

Document and test disaster recovery plan regularly, and

update as needed.

7.2 Restoring.


Identify hardware replacements.

Identify hot and cold sites.

Implement disaster recovery plan.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

network is nothing without being able to assure its users

that their data is being routinely backed up and safely stored away so that in
the event it is somehow deleted and required back, it can be restored. The
whole backup process is very elementaryit sounds easy. But establishing a
reliable backup plan and then getting it implemented and routinely managing it may be a whole different story.
Larger companies with big networks might have people whose sole
responsibility is making sure that the daily backups are happening and that
theyre reliable.
This chapter is about the concepts behind backups and their converse, restorations. Its also about a topic that goes hand in hand with backups
disaster recovery.

For complete coverage of objectives 3.1 and 4.1, please also see Chapter 7. For
complete coverage of objective 5.1, please also see Chapter 8. For complete
coverage of objective 7.1, please also see Chapters 1 and 7. For complete coverage of objective 7.2, please also see Chapter 1.

Why Do We Do Backups?
What we call the beginning is often the end. And to make an end is to make
a beginning. The end is where we start from. T. S. Eliot, Four Quarters

Truer words have not been spokenat least when we consider backups
that is. Ask yourself this question: Why would you want to back a file up in
the first place? The answer, of course, is so that if its ever needed again you

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Why Do We Do Backups?

503

can whip out your copy of it and restore it. So in reality we do backups
because we may have to perform a restoration operation someday.
But neophyte administrators dont seem to get this concept. They think
that we do backups just in case things crater in and we have to utilize them.
Truth is, most backup administrators are involved in almost daily restoration
operations. Here are some real-life examples for you to consider:


A user, working on a shared network directory whose file permissions

have been set for everyone to have full access, drags an entire folder
from one place to another, thus effectively removing the folder from
view. Other users can no longer see the folder, even though its
nearbyperhaps hiding underneath a different folder in the share
and so they call the administration team for a full restoration of the
folder. What else are they to think? The folders gone, its required,
and therefore someone on the admin team must do a restoration.

A user working on a mission-critical spreadsheet (having not hit the

Save key in the four and a half hours shes been working) walks away
from her PC to get a cup of coffee. At the same time the power in her
section of the building goes out, takes her PC out, and with it, the
spreadsheet. She calls the server admin team for a restoration.

Developers working on Web page Java code have accidentally overwritten several files in a network folderfiles that were working at one time
and now are not. The supervisor calls the admin team for a restoration.

Which of these scenarios can you assist with? If you answered all three,
then theres a problem! No server admin could assist with the middle bullet
problem. The person working on the spreadsheet has not hit the Save command
key yet and so her data has not been saved to disk. And, even if it were saved to
disk, the backups typically dont run till that evening, so there is probably very
little that could be restored to get her document back to where it was just prior
to the outage. But I include this example because this is how some users think of
backups. They have a tendency to think were out there constantly backing files
upjust in case.
So, how could we help out in the above situations? Well, the middle one we
know is a problem. But we could certainly help the user by checking to see if the
spreadsheet software shes using has an autobackup feature. Some office automation products allow you to turn on a feature whereby the document youre
working on is periodically written to a backup copy. If something goes wrong or
the computer goes down, when it comes back up you can simply open the application you were working in and the document will either reappear or youll see
it in the list of recently saved items. So theres hope, even for this hapless user.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

504

Chapter 12

Backup, Restoration, and Disaster Recovery

The other two scenarios imply that youve been routinely saving data off
to some sort of alternate media and that youre trained to understand how
to restore the missing data back to its former state once youre notified that
you need to do so. Thats what this chapter is about.

Backup Methodologies
So, when were planning our backup strategy, how do we begin? Intuitively,
we should realize that we need two different components to make this kind
of thing happen: hardware to give the backed-up data a place to live, and
software that can help us with the backup operations.
As you might imagine, you can really get crazy with your backup design
and spend thousands (or even tens or hundreds of thousands) of dollars. And
that may be necessary, but before you go that far, it might be beneficial to
examine some more cost-effective alternatives.
A backup is simply a copy operation, is it not? When you install a piece of
software on your computer, perhaps youve noticed that some installation
programs are smart enough to temporarily copy the files they are going to
overwrite to a temporary directory for safekeeping. That way, if something
goes wrong with the installation, or you want to uninstall it, the files are
present and ready to put back. Could we adopt this scenario for our backups?
Lets discuss.
Lacking any special hardware or software, its entirely conceivable that
you could set up a scenario in which you whipped up a little batch file that,
once a night, copied the contents of a critical folder to another folder for
redundancys sake. Figure 12.1 shows this. Lets begin to ask ourselves some
questions about this paradigm.
FIGURE 12.1

A poor persons backup program

Batch file

Data

Backup

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Why Do We Do Backups?

505

First of all, is there a single point of failure (SPOF)? Well, it depends on where
you place that backup copy of the data, doesnt it? If your batch file places the
backup data on the same disk as the original, then you have certainly introduced
a SPOF because, if that disk goes out, both copies go with it.
Well then, how about if you had the batch file copy the data from one disk
to another disk? Thats great unless both disks reside on the same computer,
isnt it? Because if they do, well then you still have a SPOF!
OK, you cry, how about if I have a RAID 5 array on that computer and
have two separate disk volumes and the batch file copies the data from one
place to the other? Well, youre getting warmer, much warmer. But theres
still a SPOF in the RAID array controller card, isnt there? Ditto for the idea
of copying the data from disk to, say, a Zip drivethe drive and its cable and
card are all SPOFs.
Heres the operating premise behind this whole backup methodology:
Data  Redundancy  Restore
The more redundancy there is to the data, the better your chances of being able
to restore it. So when we consider servers and backup scenarios, we begin to
think in terms of hardware to help us with the redundancy and software to
manage the backups. Lets segue into these two highly important topics one at
a time.

Hardware
More than likely, the media that you use to perform your backups will be
some sort of tape. Any of the tape media that you decide to use will require
a specific tape drive. There are several different choices of tape, but the most
common ones today are 4mm Digital Audio Tape (DAT), 8mm Helical
Scan, and Digital Linear Tape (DLT). Youll shop for a tape drive that meets
your needs and then purchase the blank media (tapes) that you need to use
in the drive. Youll also purchase some cleaning tapes so as to perform an
occasional cleaning of the drive. There are three basic kinds of tape drive/
media combinations that are available to admins:
4mm DAT DAT audiotape never really took off in the home entertainment system market, though it got some, ahem, play in the recording
studio environment. DATs claim to fame is that the data being written
to tape is digital and hence practically error-free. DAT really found a
home in the backup market, however. The 4mm DAT tape is in use in
backup drives all over the world and is one of the backup workhorses

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

506

Chapter 12

Backup, Restoration, and Disaster Recovery

for server admins. Problem is, its really slow. So if youve got a lot of
data to back up, you either have to put a bunch of DAT backup drives
on the network to facilitate multiple backups, or look at faster equipment. The 4mm part comes from the tapes width. You can copy anywhere from 224 gigabytes (GB) to one of these tapes and expect maybe
20 megabytes (MB) per minute data transfer speeds.

Note that any of the drives you buy specifically for servers will, in all probability, be SCSI-based. Which means that youd better take a look at the back of
your server first so as to ascertain what kind of SCSI attachment youll need.
Youll also usually have the choice of purchasing internal or external drives. I
much prefer external because you dont have to hassle with them if you need
to disconnect them for repairs. Youll usually have to purchase an external
cable for external drives. External drives are typically more expensive than
internal.

8mm Helical Scan Using the same kind of technology as your VCR at
home, the 8mm Helical Scan proves to be a reliable backup drive for servers. The tape is wider and holds more data, 540GB. You can realistically
expect data transfer speeds of 2060MB per minute when using a tape
drive such as this.
DLT Digital Linear Tape is the choice to make in larger networks that have
oodles of data to back up. The tape is fast and it holds lots of data. Its very
reliable. You buy DLT drives and tapes by their capacity rating: 20GB,
40GB, 70GB, etc. Typically the model number of the DLT drive youre buying will reveal its capacity. For example, if youre buying an Acme DLT7000,
you can generally assume that the 7000 number implies that the drive is a
70GB. This may not always be the case but Ive found it to be a pretty good
rule of thumb. You can expect somewhere around 120300MB per minute
of data transfer between the server and the tape drive using DLT. However,
if youve got a DLT mounted on a server and youre trying to back up data
from another server on the network, throughput might not be as robustits
relative to the speed of your network, of course.
Mammoth A fourth type of tape drive that we havent covered is
Exabytes Mammoth drive (www.exabyte.com). Data transfer rates are
very high, relative to the other listed tape drive types, around 120MB

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Why Do We Do Backups?

507

per minute. Data capacity isnt quite as high as DLT, so theres a

tradeoff in terms of speed versus the amount of data a tape can hold.
Because this is a newer technology, expect prices to be a bit higher than
older, better-established products.
With all of these data transfer speeds, keep in mind that everythings relative.
Speeds are determined not only by the rating of the backup device, but also by
the speed of the computer itself and, if youre copying across the network to
tape, the speed of the network. Just because youve got a smokin tape drive unit
doesnt mean youll get maximum throughput, especially if you hook it up to a
Pentium II 233 thats already overloaded! I like to use a Latin phrase when I
describe scenarios such as this. You must consider the computer in toto when
youre deciding about how able a box is to handle a given operation. For example, if youre running an older version of SCSI and youve got a high-speed tape
drive connected through an adapter to this slower SCSI, dont expect throughput miracles.
Note that you can buy single tape drive devices, or you can get very sophisticated in the device that you buy and can host several tape devices all at once.
If you get into the big-time tape devices, you can even have silo-type affairs with
a robotic arm that can manage hundreds of tapes. There are two basic types of
big devices like this: automated tape libraries and autoloaders. Essentially the
difference between the two is size and the number of tape drives in the system.
An autoloader takes up very little room, holds ten to twelve tapes, and usually
has only one drive. Its use is for sites where the backups span more than one tape
and you want to get away from a person having to be around to change a tape.
A tape library has multiple drives in it (usually around ten), and thus you can
manage multiple backup jobs at once.
There are many players in the tape drive market today and most of them
provide reliable products. Some of the bigger vendors are companies such as
Hewlett-Packard, Storage Tek, Compaq, Dell, IBM, and Exabyte. Pictured
in Figures 12.2 and 12.3 are the Storage Tek 9710 tape library system and
the Exabyte 110-L autoloader. These are just examplesthere are lots of
great companies and good hardware to pick from.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

508

Chapter 12

Backup, Restoration, and Disaster Recovery

FIGURE 12.2

The Storage Tek 9710 tape library

FIGURE 12.3

The Exabyte 110-L autoloader

Software
To go along with your new tape backup hardware, you need to select some
tape backup software so you can begin backing up your servers. Just as there
is quite a bit of diversity in the backup hardware arena, so it is with backup
software. In order to make a well qualified decision, you must make some
elementary determinations, as listed below:
What are your backup needs? Do you need to back up the entire server,
including the OS? If you have the OS installation CDs, then perhaps your

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Why Do We Do Backups?

509

backup scope doesnt need to be as all encompassing as backing up the

entire system. In that case you may find that you dont need an expensive
tape backup software product, and you can get by with a solution that has
less bells and whistles.
Determining the backup scope becomes a high priority because it drives
decisions about the amount of media space you require and the speed of
the backup device.
Do you have special backup requirements? Some specialized applications, particularly database applications such as Oracle and Microsoft
Exchange, may require that you purchase, at an additional cost, an
add-on module to allow you to successfully back up the application.
Pricing varies, but essentially you can count on the fact that youll need
to buy a module for each backup server that you put into place in your
environment.
There are some workarounds with some database applications. Applications
such as Oracle and Microsoft SQL Server allow you to perform an internal
backup that essentially copies the contents of the database (including its
transaction logs) to a flat file on one of the database servers disks. Then,
when you kick off your regular nightly backups, your backup software
swings by and backs up the flat file instead of the live database. You can have
great success with a method such as this provided you have a DBA who can
set up the operation and who is aware of how this methodology works.
Will you require dedicated backup servers? In smaller shops of, say, one
or two servers, you might be able to get away with one of the servers running
the backup operation while maintaining other operations as well. Certainly
the servers OS can handle such a scenario. But theres a point of diminishing
returns where it becomes necessary for you to install a dedicated backup
server, with its own set of backup tape drives and some software. This is,
without a doubt, the norm in most networks.
What will be the placement of your backup servers? In lots of networks
the administrators will put into place a server that is responsible for backups.
As a general rule of thumb, you dont want to pile numerous applications
onto a single server if you can at all avoid it. Diversifying your load onto dedicated servers helps because you can then optimize each server for its given
application and you dont take down all applications if a single server fails.
Having dedicated backup servers is a plus, but you need to consider
geographic separation in your network. For example, if you have a

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

510

Chapter 12

Backup, Restoration, and Disaster Recovery

campus in Chicago and one in Minneapolis, the two of which are connected by a moderate-speed wide area network (WAN) circuit, it may
not be prudent to install only one backup server if there is a need for
backups at both sites. As you begin to drag backup data across the
WAN link, youll see a very definite slowdown in the total throughput
of the backups and you may even find that youre not able to finish
backing up everything in a single night.
By placing backup servers at both your Chicago and Minneapolis locations, you take the WAN circuit out of the picture. We like to say that
youre performing backups locally, meaning that youre not obtaining
backups from servers on the other side of the wire.
Who will manage the backups? Tape backup systems are far from
set it and forget it. You must have at least one person whose job
includes the daily checking of the backups to make sure theyre running OK. This persons duties will involve reading the logs that were
generated as the backups ran, changing tapes, labeling and storing
tapes, and other operations pertaining to the smooth running of the
backups.
Do you have disparate systems that need to be backed up? In some
networks you have UNIX, Linux, Windows, NetWare, and other systems
all running together in some degree of harmony. Its possible, depending
on the network, that a single set of backup administrators are responsible
for maintaining backups on all the systems and, in fact, might be required
to back up all the systems onto a single set of tapes! All of this means that
the administrators must have a robust understanding of each of the Network Operating Systems (NOS) involved, and also of the connectivity
thats required to connect the system with the backup software for proper
backing up. In some cases an add-on module may need to be purchased to
go along with your regular backup software so that you can back up these
disparate systems.
Often, when setting up backups, you can start off very simply and use the
backup software that came with the NOS. My experience is predominantly
in the Windows NT and Windows 2000 (W2K) world so Ill use the W2K
software as my example. Keep in mind that other network operating systems
might have their own backup solution or might require you to purchase
third-party software.
In the W2K case, a program called NTBACKUP.EXE comes with the software.
Its not very robust, meaning that you cant purchase add-on modules for it or
do anything fancy, but if you ordered a server with a tape drive attached to it and

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Why Do We Do Backups?

511

youre going to run Windows 2000, you could easily use NTBACKUP.EXE to handle the backups on that computer and youd have very few issues. You could
even back up your Microsoft Exchange installation with NTBACKUP.EXE. There
are lots of shops out there that use NTBACKUP.EXE.
But lets say that your shop is bigger than that. Youve got a dozen servers,
for example. If you ordered all 12 servers with a tape backup drive in them
and you planned on using NTBACKUP.EXE, then your backup program might
grow arbitrarily large. While its possible for NTBACKUP.EXE to back up
other servers on the network, you might have a throughput problem or a
media space issue, so you might decide that 6 out of the 12 boxes will now
do nightly backups. Thats great except youre changing 6 tapes and reading
6 sets of backup logs each morning. There has to be a better way!
If you purchase tape backup software and install it on a dedicated backup
server whose tape drives have been sized according to your networks
backup requirements, then you have a situation where youre managing
things from a single system, you have a single set of logs to read, and a single
set of tapes to manage. Sounds very easy, doesnt it? But managing backups
is an operation that will require a great bit of your or someone elses time on
your network. You cannot neglect the routine care and feeding of your
backup installation.

If youre new to server administration, perhaps not having even gotten a job
in the industry yet, please dont be surprised or discouraged to find that your
first job involves managing the networks backup system. Oftentimes this job
is so tedious, methodical, and unchallenging that its left to junior admins to
accomplish. Yet there is no more important job than that of backing up the
network.

There are many different vendors of high-quality backup software. Allow

me to list some of the better-known backup software programs, keeping in
mind that the field is rife with competitors and you have lots of choice.
Veritas and Veritas Backup Exec A company called Seagate used to
have a backup software product called Backup Exec. But another company, Veritas (www.veritas.com), purchased Backup Exec and now
offers it as part of its product offerings, as well as its own Veritas backup
program. Veritas is well known, has been a leader in the backup industry
for a long time, and is a company you should look at when deciding about
your backup software choice.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

512

Chapter 12

Backup, Restoration, and Disaster Recovery

Computer Associates ArcServe Computer Associates (CA)

(www.cai.com/arcserve) has a tape backup program called ArcServe.
This product too has been in the industry a long time and is well known
and well understood. Computer Associates is another good company to
consider when making backup software decisions.
Syncsort Syncsort (www.syncsort.com) is a company that has a very
diversified backup software offering. Theres something for everybody,
including mainframe computers, and you would want to include this
company in your backup software research.
Legato Legato Systems (www.legato.com) has a tape backup software
offering that appeals primarily to NT and NetWare audiences. Legato is
highly thought of in the industry, and some time spent researching their
offerings is important.
This lists the leviathans in the server backup software industry, but there
are certainly other smaller fish in the pond. When you begin to research what
kind of tape backup software you want to utilize in your network, take note
of some important considerations:


What kind of support will I get with this software? Can I live with 9:00
A.M.6:00 P.M. PST support, or do I need something more realistic?

Do I have to pay for support for this product? Some companies charge
money for additional levels of technical support for their products.
Trust me on this, you will be calling the support desk for help from
time to time.

Is there a body of knowledge about this company out in the world?

Nothing helps you be a smarter, better server administrator more than
knowing others who are in the same boat with you. If you pick littleknown software, expect very few people to understand what youre
talking about when you try to communicate your problems to them.

Does this software product play in the sandbox with all of my NOSs?
Oftentimes its not advisable to go down a road where youre providing one
software product for one platform and a different product for another. Try
to stick with one brand of software in multiple NOS scenarios.

Are there drivers available for my NOS and my chosen backup device?
Its highly important to make sure that the software youre buying has
drivers that come with it that allow it to talk to the backup device youve
purchased. Its also wise to upgrade the drivers after youve installed the

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Why Do We Do Backups?

513

software and gotten it working with the device. Upgraded drivers often
contain fixes for known problems with previous device driver versions.


Will I need training on this software? Some backup software is very

complex. You may well need to invest another couple thousand dollars
just to send yourself to a class or two so you can learn how to properly
drive the software. Many admins refuse to get training on a tough piece
of software, content themselves with the softwares online help, and
then wonder why the system breaks from time to time. Chances are very
likely that the software is misconfigured. Duh.

Backup Security
You also have to plan on having a place to securely store your backup tapes
once youve begun a routine backup program (which well discuss in the next
section of this chapter). There are two or three different things you need to
think about when considering what youre going to do with all of the tapes
that begin to accumulate once the program has gotten going.
First of all, its not wise to keep your backup tapes all in one place. Think
about it. Whats the purpose of a backup tape anyway? To restore in case the
files need to be put back. Well, what happens if, God forbid, the building
burns down and your tapes burn down with it? You can buy another server,
but you cant buy the data thats on those tapes, can you? So its prudent
(and commonsensical) to somehow move a block of your most recent tapes
offsite for safekeeping.
There are a variety of methods for doing this. Ive seen everythingfrom
one man who put the backup tape in his backpack and drove it home (on his
bike) with him to his house each night, to companies that pay for a regular
service that comes to your door, picks up your tapes, and takes them to a
secure offsite storage location, and everything in between. This decision is up
to you and depends largely on the size of the backup operation and your need
for the security of the data thats on those tapes. But you can be sure that
when you consider backups, youd better plan on some sort of offsite storage
simply as a disaster recovery option if for no other reason.
Lets say that you come up with a solution where you rotate last weeks
backup tapes offsite and keep previous weeks tapes onsite in the server room.
Provided the data on the tape is any good, that scenario would provide a safe bet
that you could restore your servers up to last weeks business if something catastrophic happened, such as the fire I described above. You might not be able to
restore any historical data (because it was on the tapes that burned with the rest

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

514

Chapter 12

Backup, Restoration, and Disaster Recovery

of the buildingthe only thing you have are last weeks tapes that are currently
offsite), but you could at least get the company up and running on last weeks
stuff. Question: Is last week OK with your CEO? My guess is that she would
probably be more satisfied with restoring the business as close to the time at
which the catastrophe occurred as is possible.
So, in addition to last weeks offsite tapes, its beneficial for you to figure
out what to do with this weeks tapes and the historical data thats on tape.
You generally want this weeks tapes to be available on quick notice in case
you need to perform a restoration, but you want them to be safe too.
You can consider a fireproof data safe for this weeks tapes. Fireproof data
safes can handle a fire for one or two hours (depending on the safes rating) at
a certain temperature. Beyond that, the data in the safe isnt guaranteed to be
protected. Nevertheless, in an enterprise that takes its data seriously, a fireproof data safe is a good thing to invest in for the security of tapes that need
to be highly available. There are some major players in the data safe realm:
Schwab, Chubb, and Fireguard are three good names in fireproof data safes.
What about the historical data? Magnetic media such as tapes have a tendency
to lose (for lack of a better word) data after theyve sat around for years. As time
wears on, Im not sure you could guarantee that a tape will be as pristine as it was
the day you took it out of the drive and put it into its permanent slot. As heat, dust,
magnetism, humidity, and who knows what else work on the tape, it becomes less
and less reliable. Thats not to say that all tapes will go bad, but it is to say that
youre playing on a wider margin as the tape gets older. So what should you do
about this situation? Consider copying the tapes to a more permanent storage
source, perhaps CDs, as they get older. You could still keep the tape around as a
backup, or simply store the CD and throw the tape away.
Backup data storage service companies are worth considering. There are
two basic iterations of these companies today: media storage companies and
Internet storage companies. With media storage companies you pay a
monthly fee for someone to come pick up your tapes and take them to a special offsite location. Then, when you need a tape, you give the company your
ID number (you wouldnt want just any old body picking up one of your
backup tapes, would you?) and they deliver the requested tape to you. Youll
pay a fee based on a sliding scale for tape requests that are required more
quickly than others. For example, if youre willing to wait 48 hours for the
tape, there will be no charge. But if you need the tape in 24 hours, youll be
charged a fee. If you need it in 4, youll be charged a higher fee, and so on.
An example of this kind of company is Iron Mountain (formerly Arcus)
(www.ironmountain.com).

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Why Do We Do Backups?

515

You can also opt to use an Internet storage company to back up your data
directly to servers located on the Internet. This sounds less secure than it actually
is. There are ways of setting up server security so that only given individuals can
access an Internet server, thus providing a way for a company to contact the
server and begin backing up data. The data probably wont stay on the server,
and will be moved to an internal location, but the Internet server provides a contact point. Using an Internet company for your backup services gets you out of
the nasty business of buying, installing, and maintaining tape backup devices
and software, but its not a suitable paradigm for larger enterprises
simply because of the enormous quantity of data that needs to be backed up.
Visit Committed to Memory (www.macbackup.com) for an example of a company thats in this business (and note that this site isnt dedicated strictly to backing up Macintosh computers!).

Theres even a Web site dedicated to researching and recommending companies that act as an Internet Backup Site (IBS). Visit http://www.dantz.com/
index.php3?SCREEN=ftpproviders#criteria for a review and recommendation of some companies that are participating in the IBS arena.

Hierarchical Storage Management (HSM)

What if you have servers that get really full with user data? You do a periodic
check of the server and there are gajillions of files out there! Some of them,
you notice, are pretty old. Should you delete them? No, youve done that
before and as soon as you did you were asked to restore them from tape.
Well, should you move them off to tape? No, because, again, as soon as you
do, youll be asked to restore them to disk.
But your servers are filling up and youve got to get rid of this old stuff!
The answer lies in a technique called Hierarchical Storage Management
(HSM). The basic idea is this: You purchase an add-on module (could also
be a stand-alone module) for your backup software. You also purchase some
form of alternate media, whether the alternate form involves Storage Area
Networks (SANs), tape libraries, or something else. Then you set up some
rules that allow the system to move old data (defined according to what you
say is old data) from the server disks to the alternate media. This data is
called near line data, meaning that its immediately available to the user,
unlike a tape restoration, but that its not as close to his fingertips as it is
when its on the server. There is a momentary lapse while the system goes to
get the file requested and delivers it to the user requesting it.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

516

Chapter 12

Backup, Restoration, and Disaster Recovery

HSM can be expensive, complicated to install, and difficult to maintain.

But its a lifesaver for offloading servers from old data thats just taking up
space (and tape backup time).

Backup Security 101

When considering backup security its important to note that most backup
software programs allow you to assign a password to the tape: If the tape
were stolen and someone tried to read the data off of it, they couldnt
because they wouldnt have the password to look at the data. This is a good
idea as long as you remember the password and dont make the password
readily accessible to anyone who shouldnt see it.
You shouldnt leave your backup tapes out in the open where anyone passing by could potentially grab them. This is a very serious security breach.
Put your backup tapes behind locked doorsperhaps in the server room or
elsewhere.
Labeling backup tapes with nebulous labels helps deter people from snooping.
Which tape would a thief rather steal: P&L Statements 2000 or P2000L?
Note that some backup software programs will automatically label the tapes
with their own homegrown label if you dont choose to override and label the
tapes with a label of your choosing. Some software also provides for the ability
to barcode labels, thus preventing curious potential thieves from understanding any of the tapes contents.

Backup Types and Rotation Methods

Once youve gotten a backup solution put together, you next need to sit
down and figure out what backup type youll use and what kind of backup
rotation method youll use. Lets start by discussing the various backup type
options that you have.

Backup Types
There are three basic backup types that most backup administrators are familiar
with. Microsofts NTBACKUP.EXE program considers two other backup types. In
order for us to delve further into this topic, we need to understand that files have

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Backup Types and Rotation Methods

517

a bit in the file header called the archive bit. Usually when a file is being worked
on, the archive bit is set to false (0 or off), meaning that no backup has been performed on the file. When a backup is performed, depending on the type of
backup being performed, the archive bit may be set to true (1 or on). This bit can
act as a flag to the backup program, allowing it to avoid backing up files whose
archive bit has been set to true.
It should also be noted at this time that if a file is being held open by an application, that file probably will be skipped by the backup operation and an error
logged into the backups logs. You can purchase software that allows backups
to take place even when a file is being held open. Most backup software vendors
sell such add-on open-file modules with their product offerings.
The most common backup types are listed as follows:
Full Backup All files that have been selected for backup are backed up,
regardless of any previous backup state, i.e., whether the archive bit has
been set. It doesnt matter whether the archive bit is on or offon a full
backup, the file will be backed up anyway (unless its being held open by
a process). Note that in NTBACKUP.EXE this operation is called a normal
backup.
Incremental Backup Incremental backups (Ill call them incrementals
for short) are funny birds. Only the files that have been created or have
changed since the last full or incremental backup will be backed up. The
archive bit is set in an incremental backup, but the files date and time
stamp is checked each time the backup is run. The system says, Hmmm,
I see that the files archive bit has been cleared, but I also see that the file
has changed since last backup and it backs it up. Incrementals are a
good way to save on the time it takes to accomplish a backup.
However, theres a caveat. In order to perform a restoration on an application or system, you may have to have several tapes, up to and including the
last full backup. Think about it this way. On night one you perform a full
backup. Night two you perform an incremental. Some of the files in a given
folder dont change; others do and theyre backed up. Night three, ditto. Day
four the system goes down and you find that you have to restore the entire
folder. Night threes tape doesnt have some of the files you need for the restorationtheyre way back on night ones tape because they havent changed
since the last full. So you have to have at least two, perhaps all three tapes to
make a complete restoration work. While you save time running your nightly
backups, the tradeoff is that restorations become more complex.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

518

Chapter 12

Backup, Restoration, and Disaster Recovery

Interestingly, NTBACKUP.EXE calls incremental backups incremental

backups. Go figure. In NTBACKUP.EXE the files are marked to indicate
that the Incremental option was used (archive bit is set).
Differential Backup A differential is slightly different than an incremental. The files that have changed or have been created since the last full
backup are backed up. However, the archive bit is not set on a differential
backup, so the next differential backup will include those files.
Copy This is a Microsoft NTBACKUP.EXE term. All of the selected files
are backed up regardless of the previous backup status but the archive bit
is not set.
Daily Another Microsoft NTBACKUP.EXE choice, daily backups indicate
that the backup operation will back up only those files that have changed
today and will not set the archive bit.
As a general practice, most administrators in small shops perform a full
backup operation each night. In larger environments where there is a lot of
data that must be backed up, a full backup is taken once a week (typically
over the weekend) and then an incremental backup is performed after that.
We must also talk about a concept called a volume set. The concept is this:
You select all of the files that youd like to back up and then you give that
grouping of files a name. You can call this grouping a set or a volume or a
volume set, but the idea is that youve got a distinct grouping of files that the
backup software knows about through a single name reference. Setting up
volume sets makes it easy to name distinct backup sets and then easily keep
track of them. You might have a set called All NT Servers, for example,
while another set might be called All NetWare Servers. Or you might have
a distinct grouping of files that are relative to specific users; for example,
Marketing team shared files or Sales projections. The volume set idea is
all about organization.
Some backup software that you buy includes some default sets and
allows you to create others. The beauty of these sets is that a history of
backups that have occurred on a given set is cataloged, allowing you to sort
through and find certain restoration information that youre seeking. Lets
say youve got a backup set called Widgets. The set consists of one full
and six incrementals a week. You need to restore a file so you go looking
at the catalog to see when this file was last backed up. Once you find the
last backup of the file, you can determine the name of the tape that the file
is on, load the tape, and perform your restore.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Backup Types and Rotation Methods

519

Some backup systems keep the catalog on the tape, while others keep the catalog
in a database on the backup server. Catalog referencing is much faster when you
dont have to read the tapes header to find out whats on it, so its to your benefit
to keep catalogs on disk instead.

Another interesting option that most backup software programs offer is the
option to verify the contents of the tape after backup. This is a handy thing to
have, but takes up a lot of time. In shops where youre barely able to complete
your backups as it is, oftentimes youre faced with the option of not verifying the
data thats on the tape. Note that with database modules there is oftentimes an
option to check and perform a consistency check on the database prior to
backup and after backup. You can select one or the other or both. Just as with
verify, a consistency check takes time, thus slowing down your backups.

Backup Rotation Methods

You also need to pick out a rotation method that youll use with your backups.
When you buy backup software, the manual that comes with the software often
has a section devoted to the various backup rotation schemes that have been
invented over the years. Its important that you understand your companys
needs and pick a rotation scheme that most closely matches the time frames that
the business runs in.
The most common backup rotation method is called grandfather, father,
son (GFS). Lets say that you know youre going to back the systems up six
nights of the week. In order to implement GFS, youll need to purchase, at a
minimum, eighteen tapes. You might want to have one or two around for a
spare (dont forget a cleaning tape too). You begin your backup with the
grandfather set, as shown in Figure 12.4.
FIGURE 12.4

The grandfather setbeginning of rotation

Grandfather

Day 1
Day 2
Day 3
Day 4
Day 5
Day 6

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

520

Chapter 12

Backup, Restoration, and Disaster Recovery

When the week is up, you put these tapes in a fireproof storage vault
somewhere on site and then begin working with the father set, as shown in
Figure 12.5. Be sure to label these tapes differently than the grandfather set
so you can differentiate between them!
FIGURE 12.5

The father setpartial rotation

Grandfather

Day 1
Day 2

Vault

Father

Day 1
Day 2

Day 3

Day 3

Day 4

Day 4

Day 5

Day 5

Day 6

Day 6

Once the father set is done, rotate the grandfather set offsite, put the
father set in the storage vault, and begin using the son set. See Figure 12.6.
FIGURE 12.6

The son set complete rotation

Father

Day 1
Day 2

Vault

Grandfather

Son

Day 2

Day 3

Day 3

Day 4

Day 4

Day 5

Day 5

Day 6

Day 6

Day 1
Day 2
Day 3
Day 4
Day 5

Offsite

Day 1

Day 6

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Disaster Recovery

521

At the next backup you bring the grandfather set back onsite, the father set
goes offsite, and the son set goes into the vault. Youre all set and youve got
a GFS rotation in play. If the sky falls in and you need to perform a restoration,
at the very worst you can fall back to the week before lasts tapes. If the tapes
in the vault are secure, you gain a weeks worth of data.
Other tape rotation schemes basically follow the same kind of thinking,
but might involve lots more tapes and thus days that can elapse between set
changes, and different times when you take things offsite. You might add a
couple more sets to the rotation so that, instead of three different sets, you
have five and can extend the time that the tapes are offsite in the vault. Tape
rotation schedules are relative to the amount of money you have to spend
and the time you can put into designing a well thought out rotation schedule.

Banking on Backup Tapes

In one place that I worked as an admin, we actually took the tapes offsite and put
them in a bank storage vault that we were renting. The problem with a bank box
such as this is that its metal. Magnetic tape media doesnt do well when in contact with metal and I always wondered about the reliability of the data on those
tapes. I think I performed one restoration off them and it worked OK, but Im not
sure we could have completely counted on that data.
Its very important to design a good offsite storage patternone that you
know all involved parties will be amenable to working withand then document the system so that people are aware of whats happening.

Disaster Recovery

What this is all leading up to is the notion of disaster recovery (DR).

Its really surprising how many companies dont have a good DR plan in
place. If some truly catastrophic event were to happen, some companies
could not survive the occurrence. Theyd actually have to go out of business
because they could not restore the companys operation back to the way it
was before the disaster occurred. Or worse, they could restore, but their
operations would be so far back in time that theyd never be able to catch up.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

522

Chapter 12

Backup, Restoration, and Disaster Recovery

Heres the basic one-line premise of DR:

Be able to resume operations within <insert maximum time allowable for
the accomplishment of the restoration> back to <insert maximum allowable
number of days that business can be restored back to>. In other words, you
might have a DR mission that says, Be able to restore the companys business
operation within four hours of the occurrence of a disaster and go back to no
later than yesterdays business.
Sound like a tough call? It is! But thats the mission of DR. Backup tape
systems dont exist simply to restore someones file that they accidentally
deleted and immediately need restored back. This might be the day-to-day
usage for these systems, but tape systems primarily exist as the first segue
into your DR scenario. Youll use backup tapes to restore your companys
business operation after a disaster has occurredno matter how small that
disaster. Following are the major high-level considerations you must think
about when setting up the 30,000-foot view of your DR plan:
Determine who the stakeholders are. So, how do we begin thinking in
disaster recovery terms? First of all, communication is the name of the game.
As an administrator, youre the one wholl be taking care of the servers, but
youre not necessarily the one who has a stake in a given system going down.
For example, suppose that you have a financial system running on a server.
The system is under your care and control, but the owners of the system are
the financial peoplethey rely on the system to work daily and continually.
In part, they are some of the stakeholders of the system. You manage it, but
they have a huge vested interest in its operation. So when we consider DR
methodologies and techniques, we should be careful to include all system
stakeholders in our conversations.
Assemble the various stakeholders together, all at once or, if there are
numerous systems, a bit at a time. Find out what the underlying concerns
are about systems and document them. Formulate some ideas as to how
you would recover a given system should the sky fall and the building you
currently occupy suddenly cease to exist. In this documentation, note things
like the systems overall importance in the ongoing operation of the company (e-mail, while hugely important, probably isnt as important as the
companys financials, for example). Document who the subject matter
experts (SMEs) of a given system are. An SME isnt necessarily someone
who is technically astutehe is someone who understands all of the subtle
nuances of the system because he has worked with it so long. Note that in
enterprise systems that fall into the domain of everybody owns the system

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Disaster Recovery

523

and the admins administer it (systems such as e-mail and authentication

servers), youre the SME and the main stakeholder.
Establish the DR plan. Once youve got some concrete ideas about
what the systems are and how their owners feel about them, and youve
prioritized the systems importance in terms of the quickness-to-restore
scale, you can formulate a high-level DR plan. I say high-level because you
dont yet know how youre going to accomplish your plan, just that you
now know what systems need to be included. Include in your documentation the stakeholders and SMEs who are a part of each system.
Develop and refine your physical action plan. Next you come up with a
plan that details physically how you will see to it that these critical systems
and their associated data are part of the DR plan. There are many techniques,
some elementary, others highly sophisticated (and expensive). For example,
you might simply opt to come up with a robust backup methodology and
routinely move your tapes offsite (along with backup copies of your installation media). Or you might invest in a system where two geographically
distant Storage Area Network (SAN) farms talk to each other over a highspeed network wire and you perform real-time mirroring of the data from
one SAN to the other. Or you might come up with something in between,
such as clustered servers for a short-term high-availability solution and a
robust backup program for long term.
Putting the whole plan into place. Finally, once youve got the plan
written and you and the stakeholders are happy with it, you take it to the
people who need to approve the plan, get the money approved, buy the
gear, and get things put into place.

Highly Available Systems versus DR

At this point, you should understand that theres a difference between a
system thats highly available, meaning that there are features in the system
that will allow it to recover from errors or faults, and a DR program. In a
high-availability scenario, youre simply adding gear to existing systems or
purchasing equipment thats already equipped with fault-tolerant gear so
that the systems installed on them stand a much better chance of staying up
and running even when a physical fault occurs. DR, on the other hand,
wants to know how youre going to get the systems back up and running
after a catastrophic occurrence. DR might include high-availability scenarios but DR is a separate notion from high-availability.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

524

Chapter 12

Backup, Restoration, and Disaster Recovery

Lets consider a scenario to illustrate what I mean. Figure 12.7 shows a

cluster installation consisting of two servers in a cluster (each of which is
called a node). Nodes require a cable between them so that each can monitor
the others heartbeat to make sure the partner is running, a connection to the
network, and a connection to the data storage device. If the primary node
fails, the secondary node detects the failure and kicks in. This operation is
called a failover. When the failed server is fixed and back online and made
primary again, the operation is called a failback. Users generally see a minor
blip on the radar screen and can continue going. Sounds simple, but clustering can be incredibly complex to implement. Some cluster systems support
more than two nodes. Note in Figure 12.7 that the SPOF lies in the single
data vault. There are workarounds to this, but in this picture were showing
only one.
FIGURE 12.7

A typical cluster server scenario

To Network

To Network

Heartbeat

Node 1

Node 2

Data vault

Now lets consider DR. If the building had a natural gas leak and
exploded in the middle of the night, how many servers would be available for
you to begin restoration operations? The answer, of course, is zero. So clustering doesnt qualify as DR, but it does qualify as high-availability. Lets
take the cluster scenario a bit further. Suppose that you had a second identical cluster setup in a different building and the two clusters were connected
together by a high-speed cable (such as fiber optic running gigabit Ethernet
or ATM OC-3) and you had some software set up to copy the contents of the
data vault to the second data vault in the other building on a regular basis.
Figure 12.8 shows this scenario.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Disaster Recovery

FIGURE 12.8

525

A DR cluster server scenario

To Network 1

To Network 1

Heartbeat 1

Node 1a

Node 2a

Data vault 1
Big fat WAN pipe
(e.g. gig-ethernet, OC-3)

Data vault 2

Heartbeat 2
Node 1b

To Network 2

Node 2b

To Network 2

A scenario such as this is, as you might imagine, incredibly expensive and
includes ongoing charges such as the monthly WAN connection fee between the
two computers, not to mention the cost of the gear. However, the question of
what to do in a disastrous situation where the companys very existence hinges
on the availability of its data is key as to whether to spend the bucks or not.

You can find contractors that specialize in assisting companies with DR plan
formulation and implementation. There are people who make a living working with this stuff and theyre very knowledgeable about the latest and greatest techniques and equipment. One of the better-known DR companies is
EMCvisit them at www.emc.com.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

526

Chapter 12

Backup, Restoration, and Disaster Recovery

The site thats the primary sitethe one doing the daily workis called
the hot site. The site that acts as the repository for the backed-up systems is
called the cold site.

Poor Persons DR
The poor persons DR involves making two fundamental decisions:


Come up with a comprehensive backup plan, rotate the tapes offsite, keep
current tapes in a fireproof place, and religiously monitor the backups for
good quality and restorability. Include a copy of installation media
(Linux server, NT, W2K, or NetWare CDs, for example) offsite.

Create a methodology that describes how you will obtain replacement

servers in the event the primary servers are unavailable.

While going through the DR exercise its a good idea to tackle the high-availability
situation with your current group of servers as well. Think redundancy. RAID 5
array controllers, backup hot-swap disks, redundant power supplies and cooling
fans, and so forth are good things to have in servers.

Testing
And now for the most important part of DRtesting. In the mainframe environment there are companies that will actually act as a DR cold site for other
companies. These companies will host a once-a-year mock disaster test to see
if everythings working as planned and the DR methodologies that have been
put in place are working.
Listen, take a cue from mainframe peopletheyve been doing things like
this a long time and theyve got the bugs worked outtest, test, and test
some more your DR plan. Set up a once- or twice-a-year DR test to make
sure its working and modify according to anything new that might impact
the plan (new servers, etc.). Provide lots of DR care and feeding.
You should test restorations from time to time as well. Its like ice-skating.
If you buy a pair of ice skates and put them on from time to time, youve
proven that you can put the skates on, but you havent proven that you can
skate. You need to get out there on the ice with your restorations, especially
database restorations, and practice doing them so that when it comes to
crunch time youre not going to look like a fool because youre having to go

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Summary

527

to a book or a technical support person for answers to very basic questions.

Three A.M. on the morning of a disaster is not the time to check out the manual to see how you go about performing a restoration.

The Dot-Com That Went Oops!

Im an alternative health nut. I like to eat healthy organic foods and try to
watch the kinds of things that I eat. As a result of that, I order quite a lot of
vitamins and herbs from a company on the Internet because theyre fast
and cheap.
The other day I was out on their Web site and was startled to find that they
were saying I had to re-input all of the data I had previously given them
because the company they were contracting their financials through had
lost their data. All of it. Think about that. Lost their data. What about tape
backups? I thought. What about DR? What about.
Its inconceivable to me that a company whose very business hinges on the
success of computers would neglect such an important thingor that their
vendor would. But it happened nonetheless.
How about your company? Got backups? Got DR? Or will you too be printing
up brochures and putting embarrassing pages on your Web site announcing
that your financials were eaten and your customers will have to start all over
again?

Summary

n this chapter we talked about backups and restores. We mentioned

that there are two parts to a backup methodology: purchasing the correct
hardware and finding backup software that will fit your needs. We also
talked about the importance of restorations. Its fine that your backups are
on tape, but are you sure how to get the data back off of tape when you need
to? Are you sure that the data on the tape is goodthat youll have no problems with restorations? Restorations are the heart of backups.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

528

Chapter 12

Backup, Restoration, and Disaster Recovery

We talked about fireproof data safes and offsite storage, and combined those
notions with rotation schedules that keep part of the tapes onsite and part offsite
so that if anything happens to your facility, youll have some backup tapes that
werent harmed. An offsite tape backup operation formulates the beginning of
a disaster recovery model and, indeed, is the model in place in most businesses.
Disaster recovery asks the question, To what level of business should I be
able to restore my companys operations in the event of a disaster? Depending
on the money you have to spend and the seriousness that is placed on DR by
managers, you could wind up spending some healthy cash to get a DR scenario
going. You could have, for example, a mirrored set of DR servers, one at your
primary location, a second set in a geographically separate location, connected
together by a high-speed wire. We talked about clustering and how it is not DR
but high-availability, and we alluded to the fact that high-availability plays into
DR schemes but also works in non-DR server designs.

Exam Essentials
Know and understand backups. Know what verifying a backup means
and how to perform a regular backup. Understand what backup rotation
schemes are and why you use them.
Be able to identify different types of backup hardware. Understand the
difference between 4mm DAT, 8mm helical scan, and DLT (and possibly
Mammoth) drives. Understand that the time to back up varies between
these different offerings and that the storage capacity varies as well.
Understand backup security. Take tapes offsite, lock tapes in vaults
onsite, assign passwords to tapes, lock backup hardware up in secure rooms.
Confirm and use offsite storage for backups. Understand why youd use
offsite storage and the various options available to you for offsite storage.
Identify hardware replacements. Understand that there may be
replacements you have to make to your current computing gear in
order to a facilitate a backup operation (such as adding a SCSI card to
the system, for example) and to provide for high-availability.
Document and test a disaster recovery plan. Be able to understand why
and how you document disaster recovery and why you periodically test it.
Know what hot and cold sites are.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Summary

529

Key Terms
Before you take the exam, be certain you are familiar with the following terms:
backup
cluster
Digital Audio Tape (DAT)
Digital Linear Tape (DLT)
disaster recovery
failback
failover
Hierarchical Storage Management (HSM)
near line
restore
Storage Area Network (SAN)
single point of failure (SPOF)

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

530

Chapter 12

Backup, Restoration, and Disaster Recovery

Review Questions
1. Which of the following are considered backup methods?
A. Copying from one disk to another
B. Copying from disk to an alternative media such as CD
C. Copying from one server to another
D. Cluster server
2. Which of the following tape drives is the fastest, in terms of data I/O?
A. 4mm DAT
B. 8mm Helical Scan
C. Digital Linear Tape (DLT)
D. Mammoth
3. Which of the following drives can hold the most data?
A. 4mm DAT
B. 8mm Helical Scan
C. Digital Linear Tape (DLT)
D. Mammoth
4. When setting up a DR plan, who are the stakeholders of the system?
A. The owners of the various systems
B. Server administration team
C. Users of the system
D. Executive management

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

531

5. What is one good way to validate your backup plan?

A. Have a second admin take a look at the plan.
B. Practice performing backups.
C. Check the backup software manual for advice.
D. Practice performing restorations.
6. Youre running a Windows NT 4 server with Microsoft Exchange

Server 5.5 loaded on the computer. You want to back up the Exchange
Information Store (IS)the actual e-mail data. What tools can you use
to accomplish this?
A. NTBACKUP.EXE.
B. Third-party backup software that includes an Exchange module.
C. Stop Exchange and perform a cold backup.
D. Use NTs EXCHBKUP.EXE utility.
7. Ferdinand is reading the nightly backup event logs and runs across

this error: Warning: Computer Server3, Folder: \Users\DB, File:

security.locks.db could not be backed up. What is the most likely
cause of an error such as this?
A. Someone deleted the file from the system.
B. The file is a database thats being held open by a person or a process.
C. Backup software doesnt have permissions to access the folder or file.
D. Server is down.
8. In a grandfather, father, son backup rotation, how many sets stay

onsite?
A. 0
B. 1
C. 2
D. 3

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

532

Chapter 12

Backup, Restoration, and Disaster Recovery

9. What does it mean to verify a tape backup?

A. Check the tapes label against the backup softwares database.
B. Check the tapes consistency.
C. Check the tapes content against the files that were tagged to be

backed up.
D. Check the tape drives polling accuracy.
10. Youre the administrator of a small network of 100 users. You have

two servers. Youre setting up your backup plan. What backup type
should you select if you want to ensure complete daily backups?
A. Normal (full)
B. Incremental
C. Daily
D. Differential
11. When would you use a differential backup type?
A. When the amount of data being backed up is too large.
B. When you have different operating systems youre backing up.
C. When you are backing up databases.
D. When the files dont change very often.
12. What can you do to prevent people from stealing a tape and somehow

finding out its contents?

A. Lock the tapes up in a safe place.
B. Password-protect the tapes through backup software.
C. Take the tapes offsite.
D. Set a logon password to get into the backup software admin program.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

533

13. Suppose that you needed to run several tape backups simultaneously.

What backup device would you pick for this?

A. Autoloader
B. Tape library
C. Tape silo
D. Auto library
14. What is the main purpose of running tape backups?
A. To have a backup copy of the files on a computer.
B. To be able to restore back to a computer any files that have been

somehow corrupted or deleted.

C. To facilitate DR capabilities.
D. To provide offsite storage of files.
15. Why would you use HSM?
A. To move old data from disk to tape
B. To retrieve old data thats been archived to tape
C. To keep server disks from growing full
D. To provide defragmenting of server data
16. What three components should a DR plan include?
A. Included systems
B. Costs
C. Stakeholders
D. Subject Matter Experts (SMEs)
17. What is the standard duration used for rotation of tape sets?
A. One week
B. Two weeks
C. Three weeks
D. Four weeks

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

534

Chapter 12

Backup, Restoration, and Disaster Recovery

18. You have a group of files that are constantly being held open during

backups and your backup software simply skips them and moves on.
What can you do to solve this problem?
A. Stop whatever it is thats holding them open during backup time.
B. Purchase an open file module for your backup software.
C. Stop all services or daemons before proceeding so youre sure

nothings holding the files.

D. Set your backup software to ignore the fact that the file open is and

back it up anyway.
19. Youre a new server administrator in a very small NT shop. You know

that you need to get backups going, but money is tight right now. One
of the servers has an internal 4mm DAT tape drive and you have some
spare tapes. What can you do?
A. Download the eval of your proposed backup software from the

Web and use till it expires.

B. Borrow a copy of the software you want to use from a friend.
C. Use NTBACKUP.EXE.
D. Copy the data to CDs.
20. Why should backups be run at night?
A. Because they consume server resources while running
B. Because users wont be able to access their files while the backups

are running
C. Because they consume network resources while running
D. Because all network services stop while they run

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Answers to Review Questions

535

Answers to Review Questions

1. A, B, C. Clustering is a high-availability option that appears as

though youve backed up the data, even though you havent

youve simply moved operations from one computer to another.
When clustering, youd only have a backup if you performed a mirroring of the data thats on the current data vault to a second vault
for safekeeping. In a nutshell, any copy operation that moves data
from one point to another is a backup.
2. D. The drives are ranked in order of their I/O speed. The Mammoth

drive is fastest, in terms of data I/O, but is proprietary to Exabyte Corporation, which may or may not be OK with you. The DLT, the design
of which is licensed to many different corporations, is second fastest.
3. C. At the current storage capacity of 70GB, the DLT is by far the one with

the most data capacity. The 8mm can hold around 710GB, the 4mm 10
25GB, and the Mammoth 14GB. Keep in mind that theres a difference in
whether you choose to compress the data with your backup software or
not. If you dont compress, the tape will hold less than if you choose to
compress.
4. A, B, D. Anyone who has a vested interest in a system should be con-

sidered a stakeholder. Stakeholders will change from system to system. Users arent typically the stakeholders, though theyre certainly
involved when a systems down. Server admins will always qualify as
a stakeholder on each system. Executive management has a vested
interest because without the systems they wont have a company!
However, executives may or may not be involved in the DR planning. They probably will be involved in okaying the purchase of the
DR gear.
5. D. Youll not have validated your backup methodologies until you can

prove that you can provide a successful restoration. This might be

harder than you think, and its important that you periodically run
through a restoration test, just to make sure you know what youre
doing at crunch time. The other suggestions are OK, but remember that
backups are only as good as your ability to restore from them.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

536

Chapter 12

Backup, Restoration, and Disaster Recovery

6. A, B, C. When you install NT, the version of NTBACKUP.EXE may or may

not be able to back up the Exchange databases (the Directory Store [DS]
is considered another Exchange database). A later version of NTBACKUP
.EXE (one that comes with Service Pack 3 or above) includes updates to
[NTBACKUP.EXE] that can back up the IS and DS. Alternately, you
could investigate to see if the backup software youve chosen supports
Exchange backups, either natively or through the purchase of an add-on
module. You could also opt to stop the Exchange service and back up the
databases, but restoration using this method is sketchy at best. There is no
such thing as EXCHBKUP in NT.
7. B. The most likely cause of such an error message (I made it purposely

nebulous so you wouldnt be able to identify any one commercially available backup software product) is that the file is a database thats being
held open by a person or process. Files can be held open toothe most
notorious one being an Outlook users personal e-mail folder file (called
a PST). Typically these error messages are not ultra important unless
you absolutely require a complete backup of that file on a routine basis.
You wouldnt get the message with answer Aif the system doesnt see
a file to back up, it wont put up the above error message. Item C is a possibility, though not one Id check first. If you had a server down, youd get
a different error message (Server 3 cannot be contacted) than the one you
currently have.
8. C. Typically youll have one set thats this weeks current set. Youll

have a second set kept onsite for quick reference and a third set thats
kept offsite. The onsite set should be kept in a fireproof vault.
9. C. When you run a verification operation within your tape backup

software, youre asking the software to double-check the contents of

whats on tape against what you requested backed up to make sure
theyre the same. Verification steps take extra time in your backup
process but they help ensure that the data you have on tape is accurate.
Note that just because the verification reports an error doesnt mean
it fixes it. It just means that an error was spotted. You still have to
correct it.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Answers to Review Questions

537

10. A. The full backup operation is the one youd select to get a complete

backup of the system each night. With two servers you probably dont
have to worry very much about too much time being taken up backing
up the systems. A daily operation backs up all files but doesnt set the
archive bit. Incrementals back up those files that have changed and
sets the archive bit on those files. Differentials back up the files that
have changed and dont set the archive bit.
11. A. Typically, differential backup operations are used when youve got

so much data to back up that you cant possibly get it all backed up in
an evening to a single system (perhaps with multiple tapes or tape
drives).
12. A, B. You can add security to your backup program by making sure your

tapes are locked at all times (preferably in a fireproof vault) and by setting
passwords on the tapes so not just anybody can look at them.
13. B. Tape libraries are designed to hold hundreds of tapes (in some cases the

tapes can even be of different media types) and house several tape drives.
Thus you can run multiple backup jobs simultaneously and archive the
tapes within the unit itself. The Storage Tek 9710 device is an example of
such a tape library. The robotic arm is fun to watch and really impresses
visitors.
14. B. Without a doubt, answer B is backups main purpose. As an admin

youll have users requesting a restoration off of tape all the livelong
day. Taking backups does provide the things talked about in items A,
C, and D, but a backups real use is for restorations.
15. A, B, C. Hierarchical Storage Management (HSM) allows for data thats

old to be archived to an alternate media (either a Storage Area Network

[SAN] or tape). When a user requests the old data, theres a momentary
lapse while the data is retrieved from the alternate media, but its still
available and avoids the hassle of having to perform a restoration for the
user. HSM solutions can be expensive and difficult to implement. HSM
keeps server disks from growing full by getting rid of old untouched files
and moving them to an alternate storage place.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

538

Chapter 12

Backup, Restoration, and Disaster Recovery

16. A, C, D. Your plan doesnt necessarily need to include costs. But it

should include the systems that are to be included, the stakeholders

involved with each system, and SMEs who might have input about a
system. DR testing information and documentation should be provided as well.
17. A. Generally youll use a tape set for one week, then rotate it out and

move in the next weeks set. Youre striving for a routine regular backup
operation.
18. B. Youll see this kind of thing happen with files that users run off and

leave open for the night or with applications that have files open but
which you need to back up. Some of these files may or may not be
important to back up. You can use an open file module for your
backup software to help you back up open files such as this.
19. C. NTBACKUP.EXE will work fine for you. It will bypass open files, so

youll need to plan for that if the files are important to back up, but for
most things it works fine. NTBACKUP.EXE wont back up databases
other than Exchanges IS and DS. Copying the data to CDs is an OK
idea as well, but is resource intensive and besides that, you already
have the tapes and backup drive available.
20. A, C. Backups are server intensive and can also be network intensive

if theyre backing up other servers on the network. Therefore its not

wise to run them during business hours (though I know of some networks whose backups run so long they stretch out into the daytime
not a good thing!).

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Chapter

13

Troubleshooting and
Problem Determination

6.1 Perform problem determination.


Use questioning techniques to determine what, how, when.

Identify contact(s) responsible for problem resolution.

Use senses to observe problem (e.g., smell of smoke,

observation of unhooked cable, etc.)

6.2 Use diagnostic hardware and software tools and utilities.



Identify common diagnostic tools across the following OS:

Microsoft Windows NT/2000, Novell NetWare, UNIX, Linux,
IBM OS/2.

Perform shut down across the following OS: Microsoft

Windows NT/2000, Novell NetWare, UNIX, Linux, IBM OS/2.

Select the appropriate tool.

Use the selected tool effectively.

Replace defective hardware components as appropriate.

Identify defective FRUs and replace with correct part.

Interpret error logs, operating system errors, health logs,

and critical events.

Use documentation from previous technician successfully.

Locate and effectively use hot tips (e.g., fixes, OS updates, Esupport, Web pages, CDs).

Gather resources to get problem solved: Identify situations

requiring call for assistance; Acquire appropriate
documentation.

Describe how to perform remote troubleshooting for a wakeon-LAN.

Describe how to perform remote troubleshooting for a

remote alert.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com


6.3 Identify bottlenecks (e.g., processor, bus transfer, I/O, disk

I/O, network I/O, memory).

6.4 Identify and correct misconfigurations and/or upgrades.

6.5 Determine if problem is hardware-, software-, or
virus-related.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

uch of a server administrators time is spent troubleshooting

problems with a system. Because youre in the troubleshooting saddle so often,
its important to hone your troubleshooting skillswhich is why the Server+
exam places such import on it.
This chapter is about basic troubleshooting skills: how to develop them
and how to enhance them. Its also about how to know when you need help,
whether that help is in the form of additional human resources or electronic
or paper sources that you can go to for assistance.

For complete coverage of objectives 6.1 and 6.2, please also see Chapter 7.

Problem Determination

Finding out why a problem on a computer is occurring can be maddening.

Theres so much involvedcode, hardware, power, the network. Its all too big.
Or is it? In order to be skilled at problem determination, its important that we
break things down into their most basic elements.

Using Questioning Techniques

Heres a pretty standard troubleshooting scenario. One of your users calls
you and says, The networks down! Youre sitting at your PC logged on
to the Windows network at the time you take this call. You find that youre
able to successfully do some simple troubleshooting things:


Run Network Neighborhood and see machines listed

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

542

Chapter 13

Troubleshooting and Problem Determination

Perform a ping test to see if the TCP/IP software is running

Ping a host or two by name to make sure name-resolution services are

running

And you can do this all while youre on the phone with your user. It doesnt
take you more than 20 seconds to figure out that shes not accurately describing
the problem. So now you need to begin with a questioning section. What do
you mean, the networks down? you might ask. It is then that she tells you that
she just tried to print a document and its not printing. Perhaps you know this
person, and her computing environment, and you happen to know that she
prints to a printer that hangs out on the network, not on her local PC. Aha!
There may be a problem there. So you quickly ping the IP address of the
printerBoom! It comes right up. Hmmm. That means that its online and has
paper in it.
Is the time being displayed down in the bottom right corner of your computer? you ask her. Shes thinking youre crazy because youre asking about the
time and shes talking about a printer. But little does she know that youre trying
to figure out where the toolbar is at on her PCits movable and some users
move the darn thing all over the place. Yes, I see the time, its 3:13, she replies.
Do you see a little icon of a printer down there as well? you ask. Yes,
I do, she replies.
Can you double-click that icon for me? Id like to find out whats going
on with the printer. She does and you see that the print job is being held for
some reason. You walk her through deleting the current print job, and then
take some more time and work with her to figure out exactly how shes printing the document she wants to print. It turns out that she has selected A4
as the paper type to use, but youre in the United States. A4 is a European
paper type and wont work well with printers that are using paper sized for
U.S. printing. And in fact, if you were to visit the printer while her print job
was in the queue, youd see that the readout on the front of the printer was
prompting you to insert A4 paper.
You walk her through the steps of changing the paper selection to regular
U.S. 812-inch by 11-inch portrait and she prints it with no further problems.
Its very important, when dealing with users, to ask questions that are
fundamental. You need to work hard to communicate with users in nontechnical language and ask the right questions. I cant tell you how many
times Ive heard the phrase, the networks down, when it wasnt the network at all; it was a server that was down, or a printer, or even the telephone system. Users just dont understand all of the complexities of a
networkall of its piecesand are likely to lead you down a bunny trail
that you wont need to go down if you ask very pointed questions.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Problem Determination

543

Observing Problems
But what about when youre troubleshooting a server or network problem?
Theres no one to questionyoure all on your lonesome. You put your keen
powers of observation to work. If the servers down, exactly how is it down?
Sometimes servers are actually fully operational, but theyre responding to
one single request and its so large that theyre completely tied up. In a situation such as this, youll be able to log on to the server, but it will be moving
extremely slowly. Clearly you have something wrong, but youre not sure
just yet what it is.
Alternatively, the server might be at a stopped state and displaying an
error message. They call this the famous Blue Screen of Death (BSOD) in the
Windows NT world, but Ive seen the same sort of thing on NetWare and
Vines servers as well. Its up to your Sherlockian observation powers to note
the nature of the error and any stop codes that may be displayed on the
screen so you can find some resources that reference them, figure out what
went wrong, and correct it.
On the other hand, suppose that the server console comes up just fine and
everything looks peachy. Ah yes, but can you ping from the server console to
another host? And can you go to another host and ping the server? This is the
way that you can test network connectivity, to make sure the box is alive and
well on the network and able to take calls.
Perhaps theres a very basic problem, like the network cable is unplugged.
But you need to be cognizant that there is not only a network cable running
from the computer to the patch panel, theres also a cable running from the
patch panel to a port on a switch, hub, or router. So youve got several places
where connectivity may have gone bad. If you cant ping the host, you know
youve got a network issue of some kind, but you dont yet know what kind
it might be: hardware or software. Precisely where should you begin checking?
Youve got the network card and its drivers, the cabling, switch port, and
name server resolution to check before you can rule anything out.

Identifying Contacts
Hang with me a bit further down this road, if you will. Suppose that the server
pings fine, both out and in, its network cable is attached, and the network card
looks OK and comes up just fine. There are no apparent problems with this
unit. What then?

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

544

Chapter 13

Troubleshooting and Problem Determination

Then its time to get another face to look at the problem. Perhaps you
need an internetworking (router) person to take a look at the network to see
if anythings happening. Or maybe you need to place a call into a support
entity to see if they have any fresh ideas for you to look at.
Its important, especially when youre new to supporting servers and
users, to establish some basic support contracts with the people that have
provided your NOS and user software solutions. You need to have a place
where you can go to ask basic questions right on through to the question that
makes the support engineer on the other end of the line scratch his head.
You can also use human networking to establish some support peers.
Dont be afraid to call a buddy who works in a different company to ask her
if shes ever run into a given problem. Youll be surprised at the kind of
knowledge that people rack up over time.
You can also resort to newsgroups for some answers. But the point is that
you need to be able to identify the places you can go for the help you need.
No person is an islanddont fall into the trap of thinking that, because
youre in the server admin shoes, you cant say, I dont know. Thats the
most foolhardy of mistakes and there are a lot of admins out there who are
that way.

Using Diagnostic Tools and Utilities

ts important, then, to begin to equip yourself with some basic diagnostic

tools for troubleshooting problems on the network or the servers. The tools
you accumulate will not only consist of cognizant tools such as the ability
to think about a network cable being unplugged, but will also have in it soft
tools and possibly even hard tools. The soft tools consist of software tools and
utilities that you become aware of and use in your troubleshooting. The hard
tools consist of hardware devices specifically formulated to assist in network
evaluation. A Fluke meter, for example (www.fluke.com), might be handy if
you found that you regularly needed to know and understand what was
happening on the physical network. Fluke meters allow you to check packets,
evaluate cables, and perform a variety of other troubleshooting functions.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Using Diagnostic Tools and Utilities

545

Discovering Diagnostic Hardware and Software

Tools and Utilities
In your server admin experience, youll run across several different types of
utilities that youll try to use in order to diagnose and resolve a given problem.
Some of the utilities will work great, others wont work at all, and others will
sort of work.
For example, when working with Microsoft software, often the company
will release a product called a Resource Kit. This kit comprises many different
utilities that have been written by different people, many of whom dont work
for Microsoft but have whipped up something that helped them and so they
send it to the company for evaluation and possible inclusion in a resource kit.
For example, if youre an e-mail administrator working with Microsoft
Exchange server, youll find that its difficult to move a users mailbox from
one site to another within a given Exchange organization. (Its a piece of cake
to move a mailbox between servers in the same site, but site-to-site moves are
more difficult.) The Exchange Resource Kit provides not one, but two utilities
that facilitate the movement of user mailboxes from one site to another.
Sometimes you have to pay to obtain a given Resource Kit, while other
times Microsoft offers them for free.
The point here isnt so much about Microsoft as it is about the concept that
there are people in the world who want to work with a given problem, run into
a snafu, sit down at their favorite programming language interface, and code up
a workaround for the problem. Youll find lots of help for different issues on the
Web. So, as a budding server admin, how do you start?
First, discover the Webs great search engines and sites. Here are some
sites and tools that I regularly use in my troubleshooting efforts:
www.askjeeves.com I know its corny, but this site is great for keying
in a short problem description and finding useful help.
www.webopedia.com This site is essentially an encyclopedia for technicians. You wont find everything you need, but youll find plenty of answers.
Dont overlook the links that are often included with the answer page.
www.swynk.com This is a highly useful site for Microsoft technoids.
Oftentimes it merely links you to a support page on the Microsoft site, but
it cuts down on the amount of surfing through hits that you have to go
through.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

546

Chapter 13

Troubleshooting and Problem Determination

www.vironix.com Go out and download the coolest search engine on the

planet, Web Ferret. This little program is so dang handy I install it wherever
I go. Install it, then simply key in the search string (in quotes) that youre looking for. Youll get lots of hits, some good, some not so good.
www.microsoft.com Microsoft knows how to program Web sites and
their main site is especially good. The Search tool allows you to define the
granularity of the site so you rule out nonsense stuff such as company
announcements and non-useful things like that.
www.microsoft.com/technet TechNet is a subscription-based service
that contains a robust knowledgebase useful for solving problems associated
with Microsoft software. You can access TechNet online. Serious admins
usually subscribe to TechNet.
www.novell.com Access Novells Web site for the knowledgebase dealing
with their software.
www.ibm.com Same as Microsoft and Novell. Use this site to access
IBMs knowledgebase.
www.apple.com Ditto for Macintosh and Apple support.
www.hp.com, www.compaq.com, www.dell.com You get the idea.
Im quite sure that readers will write me and provide their own troubleshooting sites that have proven effective in the past. Keep in mind that sites change
they come and goso some of the sites I list here may be gone tomorrow.
(Microsoft, IBM, Compaq, Dell, HP, Novell, and Apple probably arent going
away anytime soon.)
Another area to look into is the idea of network monitoring software.
Careful here. You can get into serious expenditure dollars if you go overboard. Software such as Hewlett-Packards (HP) OpenView, IBM Tivoli,
CAs Unicenter TNG, and others can provide some semblance of network
monitoring. For example, even the most basic of HP OpenView Network
Node Manager (NNM) software packages can go out, check the MIBs on the
various devices on your network, and report back with a very basic health
statuswith no add-ons required. But the package is nearly $10,000 by the
time you add a dedicated server. On a 100-node networkwell, taint
worth it, as Fibber McGee used to say.
What about hardware tools? Well, we alluded earlier to the Fluke meter
(Fluke isnt the only name in the business, by the way; there are others).

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Using Diagnostic Tools and Utilities

547

Should you consider having a laptop that has some sniffing software loaded
on it for the purpose of periodically sniffing the network? Personally, I
would leave this kind of thing to internetworking experts whose job it is to
understand the nuances of the traffic flowing over the wire. Internetworkers
will likely have a dedicated Trivial File Transfer Protocol (TFTP) box sitting
around that allows for the updating of firmware code on the routers and
switches in the network.
All that highfalutin stuff being said, here are some very reasonable hardware things that you should be looking at in your troubleshooting efforts:
NICs Most network interface cards (NICs) today come equipped for either
10 megabits per second (10Base-T) or 100 megabits per second (100Base-T).
As a consequence of that, there will be a 10 and a 100 LED on the NIC (along
with an activity LED) that you can view from the backside of the computer.
Believe it or not, these LEDs are quite useful in figuring out whats going on
with your servers participation on the network. You think the NIC is set for
100, but the 10s light is flashing.
You encounter a problem when youve got a switch, whose port the server
is connected to, which is hardwired for 100Base-T full duplex. The NIC
is at 10, the switch port is at 100things are very confused. As a general
rule of thumb, switch ports should exactly match the NICs settings and
you should never go with autodetect on either the NIC or the switch port.
Ive found that autodetect tends to select the least common denominator:
Everything seems to default to 10Base-T half duplex and your server runs
like a tortoise. You usually cannot tell from LEDs if a switch port is set for
10/100 or autodetect.
Server Interfaces Servers today are shipping with readout interfaces that
describe basic characteristics about the servers health. For example, on a
typical server boot, youll see the readout on the front of the server panel
telling you whats going on as it boots. The text wont be verboseyoull
typically see messages such as booting from CMOS, but at least youll
have a clue. When a servers got an issue, you might be able to view the
problem from the servers readout. On some servers the readout even has
an elementary menu system that you can move through to assess basic
server health. Printers have a similar readout.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

548

Chapter 13

Troubleshooting and Problem Determination

From Whence Cometh the Beep?

Clara Peller became famous for her Wendys ad in which she exclaimed to
a clerk in a fast-food restaurant that wasnt a Wendys, Wheres the beef?
In one situation I was in we had to ask, Wheres the beep? We had a server
that was beeping. It was a regular beep. Monotonous. Problem was, it was
beeping in a veritable sea of servers so it was difficult to find. All the servers
seemed to be running OK, the logs looked fine, so it didnt seem to be any
big deal. But that infernal beeping! It wasnt too long before the beeping
finally got too aggravating and we set about finding out which server was
making the racket. We found the server. The power LED was red instead of
its customary amber but there was no information on the servers readout.
We found that it was a server with two power supplies, one of which didnt
have a power cord. When we plugged in an additional power cord, the
server stopped beeping and the power LED turned amber again.

Common Diagnostic Tools

Regardless of the Network Operating System (NOS), you have some tools
that are commonly in use. Some are hardware-based, others are found in
software. Here they are:
RAID Array Controller Card BIOS When you purchase a server with a
hardware-based RAID array controller card, regardless of the NOS
youre going to put on it, there will be a program that you can access as
the computer boots in order to adjust the cards settings. At boot time,
youll see a message come up that says something like, Hit Ctrl-M to
adjust BIOS settings. This is the RAID cards firmware announcing to
you that now is the time to hit Ctrl-M if you want to see what the cards
configuration is set for.
Heres the deal. Its possible that something went wrong with a drive and,
for whatever reason, the configuration on the RAID card changed. I ran
into this a few years ago. I was working with a computer installing a
Mylex RAID adapter in it. The machine would boot up OK, but after
reboot it would not. I found that the Mylex card was showing one of the
drives as DED (dead). I had to use the Mylex firmware to tell the card that
the drive was OKshow it ONL (online).

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Using Diagnostic Tools and Utilities

549

Server BIOS Just as you may have to adjust a RAID cards BIOS, so you
also have a server BIOS that may need adjusting. Two of the more interesting
and common problems you can encounter that show up as errors on the software side of the server but are adjusted on the BIOS side are the automatic
PCI IRQ feature and the default enabling of COM1, COM2, and LPT1. Both
of these features have to do with hogging IRQs that a different component in
the system needs. A server whose PCI slots are automatically set to grab an
IRQ number when a device is inserted in the bus can be really tricky to troubleshoot. Lets say that you have a NIC that youve hardwired for IRQ 5 and
now youve installed a PCI sound card. Its highly possible that the sound
card will use IRQ5 because all other IRQs are in use, even though youve got
a NIC sitting out there on IRQ 5. There may be a COM port problem with
a modem and a NIC, for example.
If you know for sure that your server will never have a modem or a printer
attached to it, you can save some IRQ difficulties by disabling the COM1,
COM2, and LTP1 ports within the system BIOS. This frees up IRQs for
other devices that you may need.
With the possible exception of a workstation being used as a server, servers
dont come equipped with USB ports.
TCP/IP Software Regardless of the NOS, you can use the TCP/IP software
thats installed to perform some basic diagnostic tests. Some, if not all, of the
TCP/IP commands are available for the NOS, though there are some deficiencies (nslookup, for example, isnt available on NT servers that dont have
certain service pack levels applied). On the whole though, the ping and
tracert commands are natively available on each NOS.

Shutdowns
Its important to understand how to perform a graceful (i.e., orderly) shutdown of a system. If you just arbitrarily walk up and hit the power button
on the computer, you stand a really great chance of harming some operation
that the computer may have been in the middle of. While software developers are getting better at writing code that accounts for abrupt shutdowns, it
is never a good idea to just abruptly power down a server without first going

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

550

Chapter 13

Troubleshooting and Problem Determination

through an orderly shutdown process. Heres how to do it with the various

network operating systems:
Microsoft Windows NT From the Windows NT 4 Server console, click
Start  Shutdown  Shutdown. Or hit Ctrl-Alt-Del and select Shutdown.
To restart the server, select Restart instead of Shutdown.
Microsoft Windows 2000 From the Windows 2000 server console,
click Start  Shutdown  Shutdown. To restart the server, select Restart
instead of Shutdown.
Novell NetWare Hit Alt+Esc to toggle to the NetWare console screen,
then type the command DOWN. To restart the NetWare server instead of
downing it, type the command RESTART SERVER instead.
UNIX To shut down UNIX computers, youll need to pipe in a grace
period (the -g switch) that users are given to get out of the system. Youll
also pipe in whether you want the hardware left operational or shut down
as well. To leave the hardware operational, youll use an init state (the -i
switch) of 0. To shut down a UNIX computer while logged in as root or
one with administrative privileges, at the prompt type shutdown -y g120 -i0. (Note the use of lowercase letters.)
Linux At a command prompt, change to the /sbin directory using the cd
/sbin command. Issue the command shutdown -h now. For a reboot, use
the command shutdown -r now. The -h switch means halt. The -r
switch means reboot. Instead of now, you can pipe in a time duration,
in seconds, to wait before shutdown. Use the -tx switch to enter a time to
wait, where x is the time duration. (Run the shutdown man page for more
details on the switches available and their usage.)
IBM OS/2 Right-click anywhere on the desktop and select Shutdown
from the context menu.
In all of the NOS cases you should never simply walk up to the box and hit
the reset button unless youre forced to (by a hard-locked screen or something
similar). While the operating systems are typically graceful enough to recover
from such an operation, you simply dont know what you have in cache at the
time you hit the reset button and what kind of data youve consequently
destroyed.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Using Diagnostic Tools and Utilities

551

Keep in mind that RAID array controller cards often have RAM chips on them
that are able to keep a substantial amount of data in cache until theres a
chance to write the information to disk. If you perform a dirty shutdown (hitting the reset button), youll likely lose the information in this chip. Some
array controller cards have an option for a battery backup module to prevent
this sort of thing from happening.

Selecting the Appropriate Tool and Using It Effectively

This is where the hard part comes inthat time in the field that you need in
order to understand exactly when you should use a given tool. For example,
suppose that a server has stopped communicating with the outside world.
What should be your first step? If you said, Give it the three-finger salute
(Ctrl-Alt-Del), you get an F for this chapter!
The first thing you should do is some elementary testing of the computer. If
physically possible, go to the computer and see if you can access its console. If
the computer is halted and theres some sort of error dump on the screen, then
you should copy down the stop code (if any) and the explanation for the stop
thats being displayed on the screen. Then you have no choice but to power it
down, wait a few seconds, and power it back up to see if itll come back up.
If the computer isnt at a stopyou can access the console and the
computer seems to be running OKthen a runaway process hasnt taken
it over and youve got some more looking to do.
Perhaps the NIC is bad in the computer. Should you immediately run to
the spare parts closet, grab a new NIC, down the box, and install it? Hardly.
Your next step is to try some very basic ping operations (provided the computers running TCP/IP) to see if you can, from the server console, ping the
server and then an outside host. If you can ping outside, can you ping from
an outside computer to the server? Youre trying to rule out cabling, switch
portsthings that are in between your server and the outside world. If you
cant ping, then theres an issue either with hardware, the TCP/IP software
(such as a service or daemon that has stopped), or perhaps the network
equipment. More drilling is needed.
But if you can ping to the outside and outside users can ping into you,
your problem lies elsewherecorrupt NIC driver, invalid NIC settings, etc.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

552

Chapter 13

Troubleshooting and Problem Determination

The point here is that your troubleshooting starts small, begins locally,
and then works outward. You can use this acronym, if you like: SLOsmall,
local, outward.

A key question to ask when troubleshooting is this: What has happened to this
computer (or on this network) recently that might have caused or contributed to
this problem? Youll be surprised, as you begin to ask this question, the kinds
of goofy things that youll discover that might interrupt a server or a network.
Coincidence reigns supreme in this arena, so dont take everything you hear at
face valuebut if a change was made and your server just augered in, chances
might be good that youve discovered the culprit.

Replacing Defective Hardware Components

Theres an unspoken axiom in the industry: Make only one change at a
time. If you are experiencing problems with a server, your first inclination
is going to be to try a lot of things all at once. This methodology might be all
right, except that you wont have any clue what fixed the problem once you
get it fixed.
Furthermore, some software, especially NOS software, can be very good
at disguising exactly what is wrong, so diagnosing a hardware problem can
become even more difficult than before.
Take, for example, a friend of mine who was busy installing her first Windows 2000 Advanced Server installation. This server was going to be a root
server, one that is at the top of the heap as far as the server farm goes. The
NOS installed just fine, with no complaints. And yet, when she rebooted the
computer after the installation was finished, it would not boot up. With
Windows 2000 you have the same Hit the F8 key to see boot options
option that you have with Windows 98 and ME. So she hit the F8 key and
had the system boot and create a log file at the same time. The boot froze
she couldnt get to the log file to read it because the computer wouldnt boot.
She then tried booting to Safe mode. No goodthe computer had a deer in
the headlights feel to itit was completely frozen. This implied that all of
the drivers were out of the picture and that it was booting to known good (if
non-robust) code.
Then she got the bright idea to boot it to VGA mode only and, voila! the
machine came up. Turns out that the Nvidia video adapter had a driver conflict

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Using Diagnostic Tools and Utilities

553

with the OS and so when the video code was read into the system, the whole
thing halted. She unplugged the video from the Nvidia card, plugged it into the
regular video input built into the motherboard, and she was up and running.
When the machine came up she deleted the Nvidia driver, set the video parameters for the motherboard video slot, and she was away to the races.
What Im trying to describe here is that hardware components often
dont surface as a problemthe software driving the hardware acts up
instead. So, especially on new systems, you really have to be thinking outside the box when youre doing some troubleshooting. You might think
that youve gotten a software installation wrong when, in reality, the
installation went fine, but theres a driver conflict.
On the other hand, some hardware problems are easy to diagnose. When
video adapters go toes up, you have no difficulty at all in diagnosing the issue
the user is either completely out of video or the screen goes to some funky color
such as orange, purple, or all gray. When network cards physically go bad (a
pretty rare occurrence), the user will just not be able to connect to the network.
There are rare occasions when NICs go bad and they begin to chatter on the network. This problem can be maddeningly difficult too, oftentimes requiring
somebody with a sniffer to detect the problem. Users typically wont have too
many issues when using a chatty NIC. Motherboards that fail take the user
down to the matno work goes on at all.

Identifying and Replacing Defective Field

Replacement Units
Recall that a Field Replacement Unit (FRU) is essentially a component that can
be replaced in toto rather than by attempting to fix one of its parts. A computers
power supply would be a good example. If a power supply went bad, youd
likely not dismantle it to see what had actually gone wrong in an attempt to fix
the problem. Well, OK, there are some people that might try such a thingbut
most admins know and understand when fixing something is beyond their comprehension. A computers motherboard is another example. You wouldnt try to
de-solder a component and then put a new one in its place. Youd be very lucky
to get things working the way that they originally did and this kind of thing just
isnt done.
Fortunately, most FRUs are very easy to find and replace. A power supply in
a computer, for example, simply abruptly stops the computer. Even though
youve got the computer plugged into a socket, it wont come up. The fan wont
spin on the back of the power supply and you know for sure its dead.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

554

Chapter 13

Troubleshooting and Problem Determination

Monitors have the same sort of problem. You really cant fix a monitor,
although its possible you may be able to take it to a specialist for fixing. One
common problem in older monitors is a device called a flyback transformer.
If a monitor has one and the thing goes bad, smoke actually comes out of the
cooling vents on the monitor and you get this sickly burning-rubber smell
wafting into the room. As a general rule of thumb, fixing a monitor is going
to be almost as expensive as replacing it and youll generally simply opt to
replace it instead. But if a monitor stops working, youll instantly know
theres something wrong, wont you?
One of the hassles with FRUs is that there are so many different part numbers
and replacement items that might pertain to the specific one you have that its
difficult to pin down exactly what an FRU is. Monitors, for example, have very
different brand names, part numbers, and so forthbut they all break down
into three standard functional categories: high-resolution graphics, low radio
frequency radiation, or normal use. Of all of the different sizes, types, models,
and styles of monitors, they all were designed to meet one of the three abovelisted needs.
This concept is also true for power supplies. Your only two concerns with
a power supply are that the new one matches the old ones output and that
the new one will fit the holes the old one came out of.
The same isnt quite as true with motherboards. Theres tremendous
diversity with these components and you need to take care that, when youre
replacing a motherboard, you replace it with the same basic feature set. Size
of the board, mounting slots, options included on the board, placement of
ports, and so forth are all common features that you have to look at.

Interpreting Error Logs, Events, and System Errors

When dealing with messages that server computers send you, youll have
three basic categories of messages:
Server Logs A good server admin will read the systems event logs on a
daily basis. A better server admin will also take steps to understand and
correct errors that are being reported by some facet of the system to the
event logs. Ive seen both kinds of adminsthose that read the logs and
blow off the errors and those that get worried about errors and try to find
out whats causing them.
Its imperative that you read and understand the systems logs. Doing so
will be a great way for you to understand how a given system ticks.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Using Diagnostic Tools and Utilities

555

Understanding the rhythm of a systems logs will help you become

attuned to log entries that look out of place and uncustomary for your
system.
Its important to note that there are times when a log entry is written to
a log file and the entry seems to point to one system when in reality the
problem lies elsewhere. For example, suppose that you have a UNIX
daemon running a given database service and you have an application
that uses the database. You see in the system logs that the application
says something on the order of, Unable to process thread so-andso. But in checking the system out, you find that the daemon isnt
running and as soon as you fire it up, the errors go away. The thing to
remember is that that systems are often intrinsically tied to one another;
if one system goes bad, another system might report an error that makes
you think theres a problem with it, when in fact the problem lies with
its partner.
Event Notifications Some systems can be enabled to page or e-mail
administrators with very brief messages that alert the admin of a problem.
Systems can also put messages up on the screen for admins to see as
theyre working at the system console. Some of these event notifications
are quite terse and require that the admin be very much in tune with the
system thats attempting to notify her. For example, one time I worked
with an enterprise fax server system that was able to send out pages and
e-mail items to admins when a problem occurred. A message might say
something like, FaxServ1 port 18 aborted incoming. Now thats not all
that explanatory, and yet, to an admin who intimately knows the system
thats contacting her, its quite clear as to what to doand whether the
message requires immediate attention or can wait.
System Errors Servers will put messages into the event logs and theyll
write messages to the screen. Servers will also put up stop codes when
there is an abrupt stop in processing. Windows NT servers have what
weve described earlier as the BSOD. When you walk up to a Windows
NT BSOD, at first your heart really races because youre not sure why the
computer went down. But if youre patient and you write down the stop
code (0x0000007a, for example) and the phraseology that goes along
with the stop code, you can then jump on the Web or on Microsofts
TechNet to see if you can find out any more detail about the stop. Your
modus operandi would be the same with another NOS, though TechNet

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

556

Chapter 13

Troubleshooting and Problem Determination

would not enter into the picture. You might, for example, contact an HP
representative if you had a Hewlett-Packard UNIX (HP/UXfondly
called Hockey Pux) box sitting at a system stop state.
System errors also occur while a system is running and continues to run.
Ive seen servers that put up virtual memory errors because too many
applications or threads of applications were running at the same time.
This is indicative of a RAM-starvation issue (and/or bad code that doesnt
clean its memory pool up when done) and needs some attention.

Using a Previous Technicians Documentation

Heres the deal with using documentation from a previous technician: Usually
there aint any. Server admins, especially admins in smaller shops where theyre
the Lone Stranger, are notoriously bad at putting together any documentation
on the systems, infrastructure, etc.
If there is documentation, you can anticipate that the person doing the
documentation might not have been the worlds best writer either. Ive seen
snippets of documentation (where the person doing the documenting felt
strongly that he needed to get information about a particular system down
on paper) that were very choppy in the writing style and conveyed little
information. Often youre left more confused than if you hadnt read any
documentation whatsoever.
Some companies hire technical writers to come in and do some thorough
system documentation. The problem with this technique is that the documentation is out of date practically as it is written because most systems are
in a continuous state of change.
Take heart though. If you take a job as a server admin, youll often find
that the previous admin has agreed to be available for some system questions
while youre busy getting up to speed on the new systems. It is at this time
that I would take a tablet and pen and write down, in outline form, the things
that the previous admin has to say.
Youll also find that the admin mightve implemented a system really badly
and so you dont give a whit about documentation. Your biggest interest will be
in rectifying the problem.
Let me give you two words to think about: change management. The
underlying concept is that you take steps to deliberately and regularly document changes that have been made to systems so that if something goes
wrong, you know what has happened and can quickly backpedal. Mainframe shops have had sophisticated change management systems in place for
years. Server admins? Thats a whole other story.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Using Diagnostic Tools and Utilities

557

Using Hot Tips, Fixes, Patches, E-Support, Web Pages, CDs

Weve already talked a lot about this above, so theres not much more treatment thats needed here. Suffice it to say that theres a lot of knowledge out
there thats yours for the taking. But the issue is that you must reach out and
look for that knowledge. For example, its great that Microsoft publishes
extensive white papers on various subjects relating to their software. But
how many of those white papers do you actually read?
I know, I knowyoure a busy person. But you know what? Microsoft,
for whatever people might think of themgood or badhas people who are
aces at project management and rolling out new systems. It is well to your
benefit to read up on Windows-based rollouts when Microsofts doing the
talking. Ditto for Oracle, Sun, HP, and others in the business. They didnt get
to be the top in their business because theyre slouches at understanding their
technology.
Keep abreast of the security and service patches and packs that are issued
for a given product. Never, never, never install a service pack on a NOS without first testing it on a test server.
You can obtain lots of Web support simply by e-mailing companies
and asking them for help with their systems. Many companies post their
knowledgebases on the Web, free or at minimal cost, for you to query
about a problem. Some companies offer technical knowledgebase information on CD. Microsofts TechNet is a good example.

Gathering Resources to Get a Problem Solved

What is a resource? Its something, or someone, that you can go to in order to
seek out and find assistance in solving a problem (or enhancing a given scenario). When youre in the throes of problem solving, you need to be very
resource-aware. What I mean by that is that often admins will encounter a problem and begin to try various things in order to somehow solve the problem
through brute force.
Folks, theres a better way. When youre in the heat of a problem thats got
a server down and youre trying desperately to figure out whats going on, the
first thing you need to do is stop and think. Write down the symptoms and any
errors that you can pull from the logs or the screen. Then go check your
resources. Start with the knowledgebase for the system or application youre
working with, if there is one. Go out to the Web site and query the knowledgebase with the symptoms youve got.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

558

Chapter 13

Troubleshooting and Problem Determination

If you have no luck, hit the newsgroups and try the Web search engines for
help. Also try asking others that are in the business if they have ever run into
what youre seeing. Oftentimes someone else will have encountered this
problem before and might have a recommendation for you. On the other
hand, Ive been in situations where Ive been told: Gosh, Ive never seen that
before! Let me know what you find out!
Some admins never get to this stage, but Im not allergic to opening support
tickets with the company that manufactured the software or hardware product
in the first place. You might have to pay for the supportthats all right. Dont
let your ego get in the way of opening up a support incident with a technical support department if youre at the end of your rope. Be advisedsome companies
have extremely good support, others are terrible at it, so youre not necessarily
guaranteed an answer to your question.
Support tickets dont work well if youve got some brand-new software
on a computer that your NOS vendor wont support or youre working with
some unknown hardware that the support department knows nothing
about. In situations like this you tend to get into finger-pointing contests
where the support technicians for the NOS point to the hardware vendor and
the hardware vendor points to the NOSyoure the one left holding the bag.

Updating Your NOS with Third-Party Add-Ons

One company I worked for was running a 50/50 combination of Novell
NetWare and Microsoft Windows NT 4. In order to facilitate some
enhanced interaction between the two systems (essentially allowing the
NT boxes to be administered by NetWare-savvy admins), the company
was advised on and decided to install NetWare Directory Services (NDS)
for NT on their domain controllers (DCs).
This immediately created problems in the support arena because any time
there was a problem with one of the DCs, Microsofts tech support pointed to
NetWare and NetWare, of course, pointed to Microsoft. On top of that, the
enhancements that were to be brought about werent meeting the expectations of the people who made the decision to install NDS in the first place.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Using Diagnostic Tools and Utilities

559

The company finally worked with some Microsoft support specialists who had
experience in removing NDS for NT to remove it from the DCs, reinstating
them back to their original System Account Master (SAM) database. But what
a headache from the support perspective!
Moral of the story? When considering upgrades, especially third-party
upgrades to known network operating systems, check, re-check, and check
again to make sure that youre getting what you think youre getting and
that therell be no support issues down the road.

Identifying Situations That Require Assistance

We talked above about admins that would sooner be roasted over hot coals
than consult others for assistanceespecially if it means calling technical
support desks for the companies that author the software or manufacture the
hardware.
In some cases, I admire that kind of Lewis-and-Clark-ish attitude. But
theres a juncture at which its wise for an admin to cut her losses and simply
call for support. Its much better for you to be able to say, I dont know
and get some help than it is to waste everybodys time trying to find an
answer to a difficult problem. You might indeed solve the problem, but what
has it cost, in terms of productivity for both yourself and the company you
work for?
So, heres the way to objectively assess the cutoff point at which you seek
assistance.
1. Determine if the problem youre worried about can wait while you

take the time to try to solve it.

2. If the problem can wait, then go ahead and try to solve it yourself. If

you know youre in over your head, get assistance right away.
3. If the problem cant wait, then do a quick check of the resources you

have at your disposal: Subscription-based knowledgebases (such as

Microsoft TechNet), Web knowledgebases, Web searches, and interfacing with other people that are in the business.
4. If you cant solve the problem within a fairly rigid time parameter, say

two hours for a server issue, then cut bait and call the technical support
people to open a trouble incident.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

560

Chapter 13

Troubleshooting and Problem Determination

What to Do If Tech Support Doesnt Have a Clue

Oftentimes the first level of technical support at a company is someone
whos new to the company, possibly even new to the business. Perhaps the
company has sent this individual through thousands of dollars worth of
training to get up to snuff in order to work at the technical support desk. But
nothing beats experience like experience.
In a case such as this, youre probably wasting your time going down the
road too far with tech support, unless you know you can get the trouble
ticket escalated to the next level where, hopefully, you have some gurus who
know what the answer is.
What do you do in a situation like this, where youre pretty sure you know
more than the tech and you dont think youll solve your problem by using
this person to help you?
Well, first of all, try to escalate the problem to a different engineer. If you
cant, then youre stuck back at square one. Go through the change management
logs to make sure that you understand whats happened to the computer during
the time you began having trouble with it. If theres nothing patently obvious
that you can quickly undo, then you hunker down for the hard fight. Next you
check the knowledgebases again, call friends, network around a little bit to see
what you can find. Run through the error logs once again. Try stopping some
services or daemons to see if the computer had an issue with one and stopped
working correctly. Check the registry to make sure key entries are valid. Check
hardware for correct operation. Try to think of all possibilities.
This isnt a good answer to your problems, but in some situations, youre
stuck simply rebuilding the system. Theres no apparent answer to the problem
and youre getting no help in solving it. Youre at crunch timethis system has
to begin working. This fish-or-cut-bait juncture may require that you decide to
take a little more time and simply rebuild the computers software installations, starting with the OS. All this while youve got 15 managers staring hotly
at your back, asking you when you think the system will be back up.

Acquiring Appropriate Documentation

Heres another interesting area to talk about. Lots of admins often take the
manuals that came with the computer, throw them in a dark closet somewhere, and then try to figure out the computer on their own. How amazingly
brilliant is that? This sounds really picayune, but its worth an admins time
to read about a given device or software application before installing it. Also

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Using Diagnostic Tools and Utilities

561

if the device or software is complex and requires extensive understanding,

its best to get some training prior to going forward with the installation.
This sounds like common sense, but youd be surprised how many people
unbox something and then just begin installing away.
Im a big advocate of going through certification training. For example,
suppose that you work in a large Compaq or Dell shop. Its well worth your
while to get some training on these computers and become certified in the
proper way to set them up. Other hardware and software companies offer
the same kind of training, by the way. Dont know how to drive SQL Server?
Then get some training, buy a third-party book from Sybex, and get certified.
Itll really help you understand certain nuances of a product that you might
not have known about had you not taken the training or read the book.
With homegrown or contracted systems, it can be difficult to find documentation. Certainly it is up to you as the caretaker of a system to try to find
any sort of documentation thats available for a proprietary system such as
this. Lacking any such documentation, at least do your successor a favor and
document the system thoroughly as you discover how it works!
Documentation may be something that you and your counterparts decide
you need so badly, but dont have enough time for, that you decide youll use
contract writers and outsource the project. I applaud that because at the very
least, youre seeing to it that some of the bases are covered. However, you
cannot expect contract technical writers to be subject matter experts with
your systemsespecially homegrown proprietary systems. So its still worth
your while to try your best to do the documentation yourself.
The Web is a very good place to check for updates to existing hardware
and software youre interested in maintaining, as well as to find resources
about problems you may be having.

Performing Remote Troubleshooting for Wake-On-LAN

Wake-on-LAN is being touted as the next great thing in server administration work. And its a really great concept. The computer you want to talk to
is powered down and youre a hundred miles away from it, connected only
by your WAN line. Wouldnt it be nice if you could somehow send a signal
to the computer to wake up and turn itself on? Thats the kind of benefit you
get out of wake-on-LAN.
But is there a possibility that the wake-on-LAN solution wouldnt work on a
computer and so youd have to go about the business of fixing it? Absolutely!

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

562

Chapter 13

Troubleshooting and Problem Determination

Whenever theres a computer and some software involved, you very much run
the risk of experiencing a problem, and wake-on-LAN is no different.
What are some things that could go wrong? Well, they start out at a very
elementary stage and work their way up. Lets go over some of the more
likely problem candidates and how youd go about solving them:


Network cable not plugged inTalk someone through plugging in a

network cable, if ones available.

NIC card not waking up PCEither the PCs client isnt loaded or the
card is bad or misconfigured. No choice but to have someone power
the computer on for you, then check the problem later.

NIC responding, computer not awakeIs the computer at a halt

screen? Perhaps the computer encountered a problem when it tried to
boot and didnt make it to red lights and siren. Youve got to call
someone to check the computer.

NIC card working and you can ping the host by namePerhaps the
client computer is a Windows-based preWindows-2000 computer
running NetBIOS. You cant ping the host by name, but it shows up
in Server Manager. Is the computer actually running TCP/IP or does it
only have a different protocol such as NetBEUI or IPX loaded? Note
that you wouldnt be able to contact the computer at all if you didnt
have that same protocol loaded in your stack as well.

Outage to building or computerIf theres no power to the computer,

it wont start up.

This problems kind of like the broken-down submarine problem. Suppose

you had a problem on the outside of your submarine. How do you go about getting out of the submarine to check the problem, when you have to stay inside the
submarine to live? With a wake-on-LAN problem, there are no utilities that you
can use to, say, telnet into the NIC to see whats going on. If the stuff at the upper
layers isnt functional, its mighty great that you were able to send a wake-up
frame to the NIC, but thats as far as it goes. Youre stuck getting someone whos
local to the computer to visit it, or you wait until you can personally get to the
computer.
There are two workarounds to this problem that you have in the server
world. First of all, put two NICs in the computer so that you can attain a

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Using Diagnostic Tools and Utilities

563

modicum of fault tolerance. If the first NIC fails, then the other one will take
over. This means that youll have to have a network cable run to each NIC
and that both connections are operational.
A second workaround involves purchasing management cards for your
servers. When you purchase a management card, you burn a port on the
motherboard, but you obtain a functionality thats very cool. You can call
the card directly by phone, walk it through its configuration settings, and
even watch the machine go through its OS boot sequence. Theyre very
sophisticated. So when a wake-on-LAN NIC fails, if youve got a management card in the server you could call the card and check to see if you can
spot why the server isnt booting.

Remotely Troubleshooting a Remote Alert

Troubleshooting a remote alert means that you must somehow have preconfigured the device so that when you receive an alert youre able to remotely
contact it for further problem diagnosis. There are freeware remote control
utilities that you can download (see http://www.uk.research.att.com/
vnc/), remote control utilities that you can purchase (such as Symantecs PCAnywhere), remote-control software that comes with systems management
software such as Microsofts Systems Management Server (SMS) and others
like it, and conventional remote operation utilities such as telnet or browserbased utilities.
The key to the remote alert situation is to have a way to access the computer or device prior to receiving the alert. An alert does you no good if you
cannot access the computer to troubleshoot it!
Some older devices support the use of the TCP/IP protocols telnet and
rexec. Ive worked with 3COM switches that allowed you to telnet into
them, select a logon type, and supply a password, then be able to view and
adjust the settings on the device. In recent years this technique has moved
from telnet to a browser-based solution, but the functionality is basically the
same. There are varying security levels that can be set on the device and the
security work needs to be done up front.
Using remote control utilities such as VNC to access a device is a wonderful
way of checking an alert out, cycling a service or daemon, or even rebooting the
box. Easy does it though because VNC presents a clear security breach on your
network if the username and password to enter are left purposely generic and
easy. Its up to you to configure VNC for optimal security. Its not a good idea

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

564

Chapter 13

Troubleshooting and Problem Determination

to allow everybody who wants the use of VNC to remotely control their computer. Think about keeping this capability well within the admin bounds.

Identifying Bottlenecks

ne of the biggest complaints that youll encounter as you administer

servers is that something seems to be slow. It takes a long time to log on, to
get information from the server, to print. You get the idea. Are users
complainers? No, theyre simply identifying an issue with your network.
Maybe theres something you can do, maybe not.
The vast majority of bottlenecks occur with RAM-starved computers.
When a computer doesnt have enough RAM to handle the amount of work
that it has to do, one of two scenarios happens. Either the computer puts off
doing work that it has been requested to do, resulting in a user being given
an error message or, more likely, the NOS uses part of the computers disks
as a second, albeit much slower, memory storage area. When a NOS is able
to perform an activity like this we call it virtual memory and the computer
is said to be paging some of its memory operations out to disk. The operation
of the computer slows to a crawl.
RAM-starved computers are easy to fix. Add RAM! Depending on the NOS,
the problem can be easy or difficult to diagnose. For example, with a Windows
NT computer, you can bring up the Task Manager, go to the Performance section of the Task Manager window, then look at the amount of memory in use
compared to the amount of actual physical memory. If the amount in use is
higher than the amount of RAM you have, then youre paging. Figure 13.1
shows the window.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Identifying Bottlenecks

FIGURE 13.1

565

The Windows NT/2000 Task Manager

Disk Input/Output (I/O) can also introduce problems if the disk activity is
high and the capability of the disk to adequately answer requests is low. It
may be possible to utilize a NOSs performance monitoring tools or add-on
performance monitoring tools to make an educated judgment about whether
the disks are labored or not. Servers running on IDE hard drives are going to
run a much higher likelihood of being overloaded than those running on
SCSI. A SCSI disks spin speed, in Revolutions Per Minute (RPM), is critical
to know. Earlier disks ran at 7,500 RPM but new SCSI disks run at 10,000.
If youve documented a disk I/O problem, you can easily change out your
slower disks to new faster ones.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

566

Chapter 13

Troubleshooting and Problem Determination

Network I/O can be much harder to diagnose. Youll need to have a pro
who knows how to sniff the network and to provide concise information
about the nature of whats making the network bottleneck. Sometimes its
the network hardware. Networks that are still on 10Base-T hubs are in the
dark ages with todays hungry network demands and youll most definitely
have bottlenecks as users are added. You can fix lots of network problems by
adding switches, updating the wiring to Category 5, updating routers, and
taking other steps to speed up the network.
Bus transfer issues happen when youre working on older computers with
slow bus transfer speeds and utilizing the computers as servers. Lots of shops
use workstation class computers as servers. This isnt a wise choice, though
often its predicated on the amount of money available for server equipment.
Todays servers have 100133MHz bus transfer speeds and can easily handle the awesome restraints that modern networking can put on them. Thats
today, but tomorrow might be a different story. Unfortunately, watching for
bottlenecks isnt a one-time considerationits an ongoing process.

Identifying and Correcting Misconfigurations

and/or Upgrades

ne of the things that can easily turn a well-running server into a

completely hosed-up computer is the addition of a service pack or security
patch or similar upgrade. Similarly, fiddling with the computers configurations when you dont know what youre doing can also create havoc on the
machine.
Its always wise to use a second test computer to install service packs or
other patches as a test before you proceed with the installation on production machines. Just because a software company strongly recommends that
you apply a given patch doesnt mean that you have to run right out and do
so. Its important to test them first to make sure theyre not going to interfere
with normal operations. Some patches may not even be required based upon
the corrections the patch makes.
You can personally goof up a server by messing around with its configuration and not writing down what youve done (change management) so you
can go back and fix it if things dont work correctly. Heres a common example. You need to free up an IRQ for a device so you go into the computers

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Determining Problem Origination

567

BIOS and turn off the COM2, COM3, and LPT1 ports. Later you find out
that an internal modem in the computer, one that used to work, was turned
off when you shut down the COM ports. Oops! The problems easy to fix,
but could be a bear to resolve if someone else is called on the problem and
isnt aware of the change you made.
The best thing you can do to cut down on these kinds of problems will be
to set up a regular communications methodology where all of the admins
communicate with one another about changes that have been made. This can
take the form of a change-management documentation sheet thats regularly
updated, weekly meetings, e-mails to one another, or whatever.

Determining Problem Origination

nfortunately, hardware problems can act in such a way that you

might think you have a software issue and vice-versa. Virus problems can
emulate software problems. So diagnosing a problems origin can be tricky.
Perhaps the best way to start is by assuring yourself that there are no
viruses on the computer. If youve got a good quality virus scanner and
youre regularly updating the virus signature file on the machine, then your
chances of encountering a virus are minimal. Good virus monitoring techniques will go a long way in keeping you from suspecting a virus as the
source of your problems.
When hardware fails, typically the symptom(s) youre encountering will be
steady and will not change. If, for example, a RAM chip has failed, most network operating systems will report the problem or might even abruptly stop.
Intermittent operation might indicate a software-oriented problem instead.
Hardware problems will generally produce a system stoppage situation.
Software problems can be much more difficult to diagnose. Conscientious
monitoring of changes to a system can at least give you an idea of what might
have happened prior to the time that the system quit working. You should be
cautious of hosting multiple applications on a server because certain key system
files might be used by different applications. Some vendors will require an older
version of a system file than others and you could well wind up with a malfunctioning software installation simply because the application is trying to use an
outdated file.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

568

Chapter 13

Troubleshooting and Problem Determination

Believe it or not, some machines should be periodically rebooted.

Applications that arent well written can occupy system memory and not
release it when the application ceases operation. Cycling the computer
frees up this memory and allows the computer to begin operation in a
fresh operating environment.
Virus-related problems can surface in wacky ways. Master boot record
viruses can emulate hardware problems. The I love you virus doesnt
affect a systems operation but it does chew up some files that a user might
need and e-mail itself to others in the users address book. A high-quality
virus monitoring program in which youre actively monitoring for viruses on
all the PCs in your shop will eliminate most if not all virus problems.

Summary

his chapter is about troubleshooting problems that might happen on

your network, on your servers, or on your users PCs. We started by talking
about problem determination. You have to ask questions about changes that
occurred on a system just prior to its problem. You should be able to identify
people or go to resources that can help you solve the problem. Sometimes a
problem will be obvious, something you can see or smell, other times youll have
to use software or hardware utilities to solve a problemyoull have to be
acquainted with your systems hardware and the NOS so that you can use any
tools that may be provided to assist you. Various network operating systems
have a variety of diagnostic tools that you can use. Its important to understand
how to use the tools and how to select the correct tool for the job.
Service packs and patches might be required to clear up some problems.
But you should be careful to test service packs or other patches prior to
installing on production systems.
Lots of software and hardware vendors provide Web-based or subscription
knowledgebases that you can utilize to solve problems. You can pay for technical support as well. Oftentimes, someone that you know might be able to help
you solve the problem.
Its important to quickly identify bottlenecks in a systems operation. If a
device has a hardware malfunction, you should be able to replace it with an
FRU. You should know what sorts of problems might occur so that you can
determine which FRUs would be best to keep on hand. A replacement power
supply for a server, for example, might be a good FRU choice to have on hand.
Reading system logs and analyzing events and alerts can be helpful in
diagnosing problems.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Summary

569

Exam Essentials
Understand good problem determination and resolution techniques Be
prepared to utilize various resources at your disposal to analyze a problem
then make a determination about how to eradicate it.
Be able to power down systems Understand the difference between a
graceful and a dirty shutdown. Be able to gracefully shut down different
systems.
Understand remote connectivity issues, such as with wake-on-LAN
adapters Know what kinds of techniques you can use to rectify problems with systems that rely on remote connectivity.
Identify bottlenecks Be able to identify and eradicate system bottlenecks.
Determine if a problem is software, hardware or virus-related A first
good step in problem determination revolves around determining
whether youre dealing with software, hardware or viruses.
Maintain system documentation Good-quality system documentation,
along with a change management system, will help you perform better
trouble diagnosis and eradication.

Key Terms
Before you take the exam, be certain you are familiar with the following
terms:
paging
RAM-starved
stop codes
virtual memory

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

570

Chapter 13

Troubleshooting and Problem Determination

Review Questions
1. Your servers hard drive light seems to be constantly on. What might

be causing the problem?

A. Slow disk I/O
B. RAM-starved
C. Bus transfer speed
D. Network speed
2. You receive an alert on your pager regarding a server thats got a

problem in a city 200 miles from you. The error message says, LPD
daemon halted. How can you remotely diagnose this problem and
hopefully correct it?
A. Rexec
B. Remote control software
C. Telnet
D. Ping
3. You have a computer in a remote site that is using a wake-on-LAN

NIC. You need the computer to wake up in order to receive a security

update patch, but the system doesnt seem to be responding. What can
you do to correct the problem?
A. Telnet into the NIC.
B. Ping the computer.
C. Use VNC to access the computer.
D. Have someone on site reboot the computer.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

571

4. Ethel, your co-administrator, had a problem on a Unix computer run-

ning Oracle and hit the Reset button to cycle the server. What kind of
shutdown is this?
A. Graceful
B. Dirty
C. Systemic
D. Reboot
5. You have an applications server that requires people from several dif-

ferent areas to be able to access the computer. Today the server went
down for no apparent reason and is sitting at a stop screen. What is
your first step in problem determination?
A. Write down the stop code and stop syntax (if any).
B. Talk to the people involved with the server to see if a change has

been made.
C. Reboot the server.
D. Check the manuals.
6. What are two diagnostic commands common to all computers running

TCP/IP?
A. Telnet
B. Rexec
C. Ping
D. Tracert
7. You have a server that is running very poorly. In discussing this with

other admins in your group, you find out that the NIC was replaced
last night. Where should you begin checking to determine why the
computer is running so slowly?
A. Check the NIC configuration.
B. Make sure the cable is connected.
C. Check the port speed on the switch the servers connected to.
D. Make sure the NIC driver is the latest and greatest version.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

572

Chapter 13

Troubleshooting and Problem Determination

8. Youre working on a client computer thats having very strange problems.

When you try to run the FDISK utility to check the disks configuration,
youre told that the master boot record (MBR) is corrupt. What could be
the problem?
A. Bad disk.
B. Virus changed master boot record (MBR).
C. BIOS is bad.
D. OS is bad.
9. You have a server with dual power supplies in it. Lately the computer

has been beeping. What could be the problem?

A. Hard drive ready to go out
B. Parity error on a DIMM
C. Power supply out
D. NIC going bad
10. You have a dedicated application server. Lately the application has

been acting somewhat erratically, even though no changes have been

made to the system for a long time. Where should you begin checking?
A. System logs
B. Network connections
C. Server I/O
D. System documentation
11. You have a server that has suddenly failed to respond. When you ping the

computer you get no response back. However, when you physically visit
the machine, you find that everything seems to be operational, except that
you cannot ping anyone from the server either. What should you suspect?
A. NIC driver corrupt.
B. NIC set at wrong speed.
C. NIC cable unplugged.
D. NIC has failed.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

573

12. What are some resources that you can use to diagnose and solve problems?
A. Friends and coworkers in the business.
B. Web-based knowledgebases.
C. System documentation.
D. Try different things in an attempt to solve the problem.
13. A server that is RAM-starved will have which of the following symptoms?
A. Very slow speed
B. Errors in event logs
C. Excessive disk activity
D. Spontaneous rebooting
14. Last night you were working on a server, changing out its NIC. Coin-

cidentally, the network team was working on a router at the same

time. When you left, the NIC seemed to be working OK. When you
came into work today, you were bombarded with complaints that no
one could reach the server. Where should you begin checking first?
A. Check with the network team to see what they did wrong.
B. Check the NIC to make sure its plugged in.
C. See if you can ping the NIC.
D. Check the speed the NIC is set for.
15. The command shutdown -h now is used by which NOS?
A. Linux
B. Microsoft Windows NT
C. UNIX
D. Novell NetWare
E. IBM OS/2

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

574

Chapter 13

Troubleshooting and Problem Determination

16. Which of the following are examples of a component for which you

should consider maintaining a Field Replacement Unit (FRU)?

A. Cooling fan
B. Power supply
C. Hard drive
D. Motherboard
17. You are just starting a new server administration job when a server

fails and the reason it failed isnt evident through conventional

troubleshooting techniques. Where should begin checking first?
A. Check previous admins documentation.
B. Check system documentation.
C. Check event logs.
D. Call some networking friends.
18. Select the utility or utilities that are common across Windows NT,

Windows 2000, OS/2, NetWare, and UNIX.

A. RAID array adapter BIOS
B. Computer BIOS
C. Certain TCP/IP commands
D. FDISK
19. Youre an administrator with a server thats down for the count. You

think youve isolated the problem to a bad power supply. Whats your
next step?
A. Replace the server.
B. Replace the power supply.
C. Check that the power supply really is bad.
D. Replace the power cord.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Review Questions

575

20. A new service patch is out for your system and the documentation on

the Web indicates that it might fix some problems youve been experiencing. What should you do first?
A. Check with others to see if theyve applied the patch.
B. Test the service patch on a comparably equipped lab server.
C. Immediately install on the server with the problem.
D. Wait for a couple of months to see how the patch works out for

others.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

576

Chapter 13

Troubleshooting and Problem Determination

Answers to Review Questions

1. A, B. A slow disk or a system that doesnt have enough RAM and is

forced to page some things out to disk can create the problem
described above. Use system diagnostics to determine which might be
the case.
2. A, B. A daemon is a UNIX service. LPD is a printing service. Youre

being told that the computer cant print. Use rexec or previously
installed remote control software to get into the computer and see if
you can restart the daemon. Telnet wont work unless there has been
code written for it for your system in order to perform remote control
activities. Ping merely tells you the server is accessible through TCP/IP.
3. B, D. Theres little you can do in a situation such as this. If the NIC is

not responding, you can find that by trying to ping the NIC (provided
youre running TCP/IP, that is). Telnetting into the NIC, even if that
were possible, would not reveal much information. Youre stuck having someone onsite cycle the computer, then rechecking.
4. B. Perhaps the computer was completely frozen and Ethel had no

choice. Thats certainly a reason for having to perform a hard shutdown. But in most circumstances, youll want to perform a graceful
shutdown on serversespecially a server thats running a database
such as Oracle.
5. A. Before you do anything else, write down the stop code, if there is one,

and the syntax of the stop message, if there is one. Youll need this information as you investigate the cause of the problem. Then you begin talking to others about what may have been done to the server. By the way,
in situations such as this, its a good idea to limit the people that are
allowed on the computer and to require people to enter changes into a
change management log.
6. A, C, D. Windows NT systems cant use rexec. Telnet might be some-

what useful for diagnosing whether you at least can access the computer or notprovided, that is, that the computer has the ability to
allow telnet operations. But ping and tracert are very good commands
for diagnosing whether the computer is answering requests and if
theres a network issue surrounding its availability.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Answers to Review Questions

577

7. A. The fact that the computer had its NIC updated last night means

that this is where you should start checking first. Chances are very
good that youll find a normal installation with the exception of the
speed at which it is set. Remember that if a NIC is set at autodetect for
its port speed and the switch port is set the same, youve very likely
defaulted down to the least common denominator10 Mb/sec, half
duplex. This setting is enough to really whack the computers I/O
capabilities. Also there may be some negotiation thrashing going on as
the NIC and the switch port continue to query each other to see what
speed the other is running at.
8. B. Some viruses can change the MBR. Were not told if this computer has

a virus scanner on it or not, but even if it did and the virus signature file
was out of date, this problem could happen anyway. The BIOS being bad
would mean the computer wouldnt be available at all. Ditto for the OS.
A bad disk would probably not allow the computer to be accessible, but
it could possibly act erratically while it was going bad.
9. C. A beep on a computer can mean a variety of thingscheck

system documentation. But with some servers youll get a beep if

one of the power supplies is either not plugged in or has failed.
Replace with an FRU.
10. A. Typically, an application written for the NOS on which it is run-

ning will post events to the systems log files. You can glean tons of
information simply by reading the logs to see if you can find out
whats wrong. If you come up blank, begin to dig into the system
documentation, or use the documentation along with the log files to
interpret whats happening.
11. C, D. Answer A is a little sketchy because NIC drivers get loaded into

memory. Even if the driver became corrupt, it would not initially be

responsible for the current failure. It might fail after you resolve the
problem and reboot, but thats then, this is now. The NICs cable is
possibly unplugged or it has failed. Note that the cable could have
failed and caused the problem as well. The wrong speed typically
wont keep the box from operating on the network, just make it very,
very slow.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

578

Chapter 13

Troubleshooting and Problem Determination

12. A, B, C. Answer D is something that you shouldnt let yourself get

involved with. Ask friends and coworkers who are familiar with systems, check knowledgebases, and consult system documentation, but
dont try to solve problems by trying one thing after another!
13. A, C. Servers that are RAM-starved are bottlenecking by having to page

data out to disk. The server will be very slow and youll see excessive disk
activity. There will not necessarily be any errors in the event logs. Spontaneous rebooting is a sign of a much larger problem.
14. B. First check to make sure the NICs operational by pinging it from

a workstation, then by pinging hosts from the server itself. Check the
NICs speed as well. You no doubt did all this last night, but when
youre working on servers late at night, its easy to get things mixed
up. Then, having ruled out that your installation is the problem, its
time for you to drill in on the network team. Pinging the NIC might
serve to shortcut you to the network team right away.
15. A. The Linux and UNIX shutdown commands are quite similar, but

the -h (halt) switch is specific to Linux.

16. A, B, D. Technically a hard drive isnt considered an FRU, though

with todays hardware-based RAID systems you could keep a cold

spare around in case the disk failed.
17. A. Begin by checking to see if the previous administrator left any doc-

umentation on this system. Perhaps there is some previous water

under the bridge that you need to be aware of. Youre told that conventional methods, i.e., checking the event logs, doesnt yield any
information. Checking the system documentation is a good idea, but
any previous admin documentation might be the best place to begin
looking.
18. A, B, C. FDISK is a DOS utility, not Linux. Linux creates a LILO par-

tition that isnt recognizable by FDISK. But answers A, B, and C have

a commonality across all the network operating systems.
19. D. Power consists of a lot of things and we often forget that the com-

puter connects through cords. If you think you have a power supply
problem, check the cord first before proceeding with problem resolution. Power cords can and do go bad.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Answers to Review Questions

579

20. B. While you should never simply haul off and install a service patch on

a production server, on the other hand youve got the potential to solve
some problems youve been having. So immediately download the patch
but test it on a lab box. Note that the lab computer will have to be comparably equipped with the hardware and apps you have on your problem
server so that you get an accurate test. Checking with others is a good
idea, but wont work if youre an early adopter. Waiting it out is also a
good idea provided you dont have an immediate need.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Server+ Glossary

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

582

Glossary

386 enhanced mode In Microsoft Windows,

the most advanced and complex of the different
operating modes, 386 enhanced mode lets Windows access the protected mode of the 80386 (or
higher) processor for extended memory management and multitasking for both Windows and
non-Windows application programs.

100Base-TX This is the specification for

operating Fast Ethernet (100Mbps) over two
pairs of Unshielded Twisted Pair Category 5
cable. The specification has a maximum cable
distance between the port and the device of 100
meters, which includes all patch cables, panels,
and connectors.

1000Base-CX The specification for running

Gigabit Ethernet over short distances on
shielded twisted pair cable.

100Base-X Generic family of cabling standards

that includes 100Base-TX, 100Base-FX, and
100Base-T4.

1000Base-LX This is one of the specifications

for running Gigabit Ethernet over fiber optic
cable. The L stands for long wavelength, which
defines the optical transceivers or laser devices.
The wavelength of the LX lasers is 1270 to
1355 nanometers (nm), which is commonly
referred to as 1350 nm.

10Base Fiber Backbone (10Base-FB) This

is the specification for operating half duplex
Ethernet (10Mbps) over long distances on two
fiber optic cables at distances up to 2000
meters.

1000Base-SX This is one of the specifications for running Gigabit Ethernet over fiber
optic cable. The S stands for short wavelength,
which defines the optical transceivers or laser
devices. The wavelength of the SX lasers is 770
to 860 nanometers (nm), which is usually
referred to as 850 nm.
1000Base-T Standard for Gigabit Ethernet
over Category 5 copper cable.
100Base-FX This is the specification for running Fast Ethernet (100Mbps) over fiber optic
cable.
100Base-T4 This is the specification for operating Fast Ethernet (100Mbps) over four pairs
of copper wire, with a signaling system that
allows Fast Ethernet to be carried over voice
grade, Unshielded Twisted Pair Cat 3 cabling.

10Base Fiber Link (10Base-FL) This is the

specification for operating Ethernet over long distances on two fiber optic cables. In half duplex
mode, 10Base-FL has a distance limitation of
2000 meters. In full duplex mode, the signal can
travel over 2000 meters.
10Base Fiber Passive (10Base-FP) This is
the specification for operating half duplex
Ethernet (10Mbps) over long distances on two
fiber optic cables at distances up to 1000
meters.
10Base2 This is otherwise known as thinnet.
This is Ethernet operating on a single 50-ohm RG
58 coaxial cable. The 10Base2 distance limitation
is 185 meters.
10Base5 This is otherwise known as
thicknet. This is Ethernet operating on a single
50-ohm 10mm thick coaxial cable. The
10Base5 distance limitation is 500 meters.
10Base-F This is the Ethernet specification
for operating on fiber optic cable.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

10Base-FL

See 10Base Fiber Link.

10Base-T This is the Ethernet specification for

operating on two pairs of 100-ohm Category 3 or
better UTP cable. 10Base-T can operate in either
half duplex or full duplex mode at distances up to
100 meters.
10Broad36 This is the specification for
operating Ethernet over a single 75-ohm
CATV broadband cable. Maximum distance
is 1800 meters in half duplex mode only.
802.3 This IEEE standard defines a bus
topology network that uses a 50-ohm coaxial
baseband cable and carries transmissions at
10Mbps. This standard groups data bits into
frames and uses the Carrier Sense Multiple
Access with Collision Detection (CSMA/CD)
cable access method to put data on the cable.
802.5 The IEEE 802.5 standard specifies a
physical star, logical ring topology that uses a
token-passing technology to put the data on the
cable. IBM developed this technology for their
mainframe and minicomputer networks. IBMs
name for it was Token Ring. The name stuck,
and any network using this type of technology
is called a Token Ring network.
80286 Also called the 286. A 16-bit microprocessor from Intel, first released in February
1982 and used by IBM in the IBM PC/AT
computer. Since then it has been used in many
other IBM-compatible computers. The 80286
uses a 16-bit data word and a 16-bit data bus,
and it uses 24 bits to address memory.
80287 Also called the 287. A floating-point
processor from Intel, designed for use with the
80286 CPU chip. When supported by application programs, a floating-point processor can

583

speed up floating-point and transcendental math

operations by 10 to 50 times. The 80287 conforms to the IEEE 754-1985 standard for binary
floating-point operations, and it is available in
clock speeds of 6, 8, 10, and 12MHz.
80386DX Also called the 80386, the 386DX,
and the 386. A full 32-bit microprocessor introduced by Intel in October 1985 and used in many
IBM and IBM-compatible computers. Available
in 16MHz, 20MHz, 25MHz, and 33MHz versions, the 80386 has a 32-bit data word, can
transfer information 32 bits at a time over the
data bus, and can use 32 bits in addressing
memory. The 80386 is equivalent to about
275,000 transistors, and can perform 6 million
instructions per second. The floating-point processor for the 80386DX is the 80387.
80386SX Also called the 386SX. A lower-cost
alternative to the 80386DX microprocessor,
80386SX was introduced by Intel in 1988. Available in 16MHz, 20MHz, 25MHz, and 33MHz
versions, the 80386SX is an 80386DX with a 16bit data bus. This design allows systems to be configured using cheaper 16-bit components, leading
to a lower overall cost. The floating-point processor for the 80386SX is the 80387SX.
80387 Also called the 387. A floating-point
processor from Intel, 80387 was designed for use
with the 80386 CPU chip. When supported by
application programs, a floating-point processor
can speed up floating-point and transcendental
math operations by 10 to 50 times. The 80387
conforms to the IEEE 754-1985 standard for
binary floating-point operations and is available
in speeds of 16, 20, 25, and 33MHz.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

584

Glossary

80486DX Also called the 486 or i486.

80486DX is a 32-bit microprocessor introduced
by Intel in April 1989. The 80486 represents the
continuing evolution of the 80386 family of
microprocessors and adds several notable features, including on-board cache, built-in floatingpoint processor, and memory management unit,
as well as certain advanced provisions for multiprocessing. Available in 25, 33, and 50MHz versions, the 80486 is equivalent to 1.25 million
transistors and can perform 20 million instructions per second.
80486DX2 Also known as the 486DX2. A
32-bit microprocessor introduced by Intel in
1992. It is functionally identical to and 100
percent compatible with the 80486DX, but it
has one major difference: The DX2 chip adds
what Intel calls speed-doubling technology
meaning that it runs twice as fast internally as
it does with components external to the chip.
For example, the DX2-50 operates at 50MHz
internally but at 25MHz while communicating with other system components,
including memory and the other chips on the
motherboard, thus maintaining its overall
system compatibility. 50MHz and 66MHz
versions of the DX2 are available. The
486DX2 contains 1.2 million transistors and
is capable of 40 million instructions per
second.
80486SX Also called the 486SX. A 32-bit
microprocessor introduced by Intel in April
1991. The 80486SX can be described as an
80486DX with the floating-point processor circuitry disabled. Available in 16MHz, 20MHz,
and 25MHz versions, the 80486SX contains
the equivalent of 1.185 million transistors and
can execute 16.5 million instructions per
second.

80487 Also called the 487. A floating-point

processor from Intel, designed for use with the
80486SX CPU chip. When supported by application programs, a floating-point processor can
speed up floating-point and transcendental
math operations by 10 to 50 times. The 80487
is essentially a 20MHz 80486 with the floatingpoint circuitry still enabled. When an 80487 is
added into the coprocessor socket of a motherboard running the 80486SX, it effectively
becomes the main processor, shutting down the
80486SX and taking over all operations. The
80487 conforms to the IEEE 754-1985 standard for binary floating-point operations.
8086 This 16-bit microprocessor from Intel
was first released in June 1978, and it is available in speeds of 4.77MHz, 8MHz, and
10MHz. The 8086 was used in a variety of
early IBM-compatible computers as well as the
IBM PS/2 Model 25 and Model 30. The 8086
uses a 16-bit data word and a 16-bit data bus.
The 8086 contains the equivalent of 29,000
transistors and can execute 0.33 million
instructions per second.
8088 This 16-bit microprocessor from Intel
was released in June 1978, and it was used in
the first IBM PC, as well as the IBM PC/XT,
Portable PC, PCjr, and a large number of IBMcompatible computers. The 8088 uses a 16-bit
data word, but transfers information along an
8-bit data bus. Available in speeds of 4.77MHz
and 8MHz, the 8088 is approximately equivalent to 29,000 transistors and can execute 0.33
million instructions per second.
8-bit bus The type of expansion bus that was
used with the original IBM PC. The bus can
transmit 8 bits at a time.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

abend Abbreviation for Abnormal End.

Novells version of the Blue Screen of Death.
Accelerated Graphics Port (AGP) bus A
type of 32-bit expansion bus that runs at
66MHz. It is a very high-speed bus that is used
primarily for video expansion cards and can
transfer data at a maximum throughput
508.6MBps.
Access Control Lists (ACLs) A list associated with a file or directory that contains information on which users can access the file or
directory, and which permissions each user can
apply.
access time The period of time that elapses
between a request for information from disk or
memory and the information arriving at the
requesting device. Memory access time refers to
the time it takes to transfer a character from
memory to or from the processor, while disk
access time refers to the time it takes to place
the read/write heads over the requested data.
Active Directory The Active Directory, a
new feature of Windows 2000, stores information about users, computers, and network
resources. The Active Directory is stored in
databases on special Windows 2000 Server
computers called Domain Controllers.
active hub A type of hub that uses electronics to amplify and clean up the signal
before it is broadcast to the other ports.
active matrix A type of liquid crystal display
that has a transistor for each pixel in the screen.

585

trol every pixel on the screen. Active-matrix

screens are characterized by high contrast, a
wide viewing angle, vivid colors, and fast
screen refresh rates, and they do not show the
streaking or shadowing that is common with
cheaper LCD technology.
active monitor The active monitor on a
Token Ring network is the first station that is
powered up. It actively monitors the ring.
actuator arm The device inside a hard disk
drive that moves the read/write heads as a
group in the fixed disk.
Adapter Load Balancing A process where
network interface cards (NICs) will work
together to make sure each card is handling an
even share of the packets coming into the
server.
Adapter Teaming Two or more network
cards servicing the same network segment and
providing increased data throughput.
address bus The internal processor bus used
for accessing memory. The width of this bus
determines how much physical memory a processor can access.
address The precise location in memory or
on disk where a piece of information is stored.
Every byte in memory and every sector on a
disk has its own unique addresses.
Administrative Tools In Windows 2000
Professional, this is the group of utilities used to
manage many common configuration and
maintenance tasks.

active-matrix screen An LCD display mechanism that uses an individual transistor to con-

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

586

Glossary

algorithm
problem.

A formula for solving a complex

allocation unit An allocation unit is a portion of the hard drive that is used by the computer when saving information to the drive.
Smaller allocation units are generally more efficient, because they result in less wasted space.
alpha Extremely early versions of computer
software are called alpha code. Alpha code is
generally incomplete and unusable, and it is
almost never released to the public.
American National Standards Institute
(ANSI) ANSI is the organization that is recognized in the United States as the body that
develops technology standards. It is a member
of the International Organization for Standardization.
analog Describes any device that represents
changing values by a continuously variable
physical property such as voltage in a circuit,
fluid pressure, liquid level, and so on. An
analog device can handle an infinite number of
values within its range.
anti-static bag A bag designed to keep static
charges from building up on the outside of a
computer component during shipping. The bag
will collect some of the charges, but does not
drain them away as ESD mats do.
anti-static wrist strap (ESD strap) A specially constructed strap worn as a preventive
measure to guard against the damages of ESD.
One end of the strap is attached to an earth
ground and the other is wrapped around the
technicians wrist.

anti-virus program An application program

you run to detect or eliminate a computer virus
or infection. Some anti-virus programs are terminate-and-stay-resident programs that can
detect suspicious activity on your computer as
it happens, while others must be run periodically as part of your normal housekeeping
activities.
Application layer The seventh, or highest,
layer in the International Organization for
Standardizations Open Systems Interconnection (ISO/OSI) model for computer-to-computer communications. This layer uses services
provided by the lower layers, but is completely
insulated from the details of the network hardware. It describes how application programs
interact with the network operating system,
including database management, e-mail, and
terminal emulation programs.
application server A process running on a
computer in a distributed network that provides all the business logic for an application
program.
arbitrated loop In Fibre Channel, an arbitrated loop is a connection that will allow up to
127 devices to share bandwidth. The devices
will be connected in a daisy-chain fashion.
ASCII Acronym for American Standard Code
for Information Interchange. A standard
coding scheme that assigns numeric values to
letters, numbers, punctuation marks, and control characters, to achieve compatibility among
different computers and peripherals.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

asynchronous Describes a type of communication that adds special signaling bits to each
end of the data. The bit at the beginning of the
information signals the start of the data and is
known as the start bit. The next few bits are the
actual data that needs to be sent. Those bits are
known as the data bits. Stop bits indicate that
the data is finished. Asynchronous communications have no timing signal.
asynchronous mode Method of communication between a SCSI controller and a SCSI
device that does not require any sort of timing
signal.
Asynchronous Transfer Mode (ATM) In
banking, an ATM is an instant cash machine.
In computing, it is a switching technology that
uses dedicated connections to manage digital
data in 53-byte cells. These cells are then transmitted over a physical medium using a digital
signal technology.
AT Attachment (ATA) ANSI group
X3T10s official name for Integrated Drive
Electronics (IDE) devices.
AT bus Another name for the ISA bus. See
also ISA.
ATA 100 IDE interface that will transfer data
at up to 100Mbps.
ATA 2

See ATA version 2.

ATA 3 ATA 3 was a minor revision to ATA

version 2.
ATA Packet Interface (ATAPI) This is the
IDE interface that works between your computer and either CD-ROM devices or tape
backup drives.

587

ATA version 2 (ATA 2) The second version

of the original IDE (ATA) specification that
allowed drive sizes of several gigabytes and
overcame the limitation of 528MB. It is also
sometimes generically known as Enhanced IDE
(EIDE).
Attached Resource Computer Network
(ARCNet) A network technology that uses a
physical star, logical ring, and token passing
access method. It is typically wired with coaxial
cable.
Attachment Unit Interface (AUI) A 15-pin
interface that connects a computers network
interface card and an Ethernet cable.
AUTOEXEC.BAT A contraction of AUTOmatically EXECuted BATch. AUTOEXEC.BAT
is a special DOS batch file, located in the root
directory of a startup disk, and it runs automatically every time the computer is started or
restarted.
auto-negotiation The process that Ethernet
network cards use to determine the speed of the
network. This speed can either be 10Mbps or
100Mbps.
auto-ranging multimeters A multimeter
that automatically sets its upper and lower
ranges depending on the input signal. These
multimeters are more difficult to damage by
choosing the wrong range setting. See also
multimeter.
auto-reconfiguration The process a Token
Ring network will use to re-create a ring if a
network card is unexpectedly removed from
the ring.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

588

Glossary

Autorun On a CD-ROM, the Autorun

option allows the CD to automatically start an
installation program or a menu screen when it
is inserted into the CD-ROM drive.
baby AT A type of motherboard form factor
where the motherboard is smaller than the original AT form factor.
backbone A transmission line that is used to
interconnect smaller network segments.
backup A duplicate copy made to be able to
recover from an accidental loss of data.
Backup Domain Controller (BDC) In Windows NT, a BDC is a copy of the Primary
Domain Controller than is used to provide
authentication to the network.
backup set
media.

A related collection of backup

backup software Program that is used to

back up a small amount of data on a floppy disk.
backup source
backed up.

The device or data being

bandwidth In communications, the difference between the highest and the lowest frequencies available for transmission in any given
range. In networking, the transmission capacity
of a computer or a communications channel
stated in megabits or megabytes per second; the
higher the number, the faster the data transmission takes place.
Bandwidth Allocation Protocol A management protocol that will control the amount of
the bandwidth that is being used by a process or
service.

Basic Hot Swap IN PCI, one of the ways

that devices can be added or replaced without
taking a server offline.
Basic Input/Output System (BIOS)
BIOS.

See

basis weight A measurement of the heaviness of paper. The number is the weight, in
pounds, of 500 11" 17" sheets of that type of
paper.
batch file File with a .bat extension that
contains other DOS commands. By typing the
name of the batch file and pressing Enter, DOS
will process all of the batch file commands, one
at a time, without need for any additional user
input.
baud rate In communications equipment, a
measurement of the number of state changes
(from 0 to 1 or vice versa) per second on an
asynchronous communications channel.
Berg connector A type of connector most
commonly used in PC floppy drive power cables;
it has four conductors arranged in a row.
beta Beta code is software that has reached
the stage where is usable and generally stable,
but it is not completely finished. Beta code is
often released to the public for testing on an as
is basis, and user comments are then used to
finish the release version of the product.
bias voltage The high-voltage charge
applied to the developing roller inside an EP
cartridge.
bidirectional Operating in two directions at
the same time.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

binary Any scheme that uses two different

states, components, conditions, or conclusions.
In mathematics, the binary (base-2) numbering
system uses combinations of the digits 0 and 1
to represent all values.
BIOS (basic input/output system) The
ROM-based software on a motherboard that
acts as a kind of interpreter between an operating system and a computers hardware.
BIOS CMOS setup program Program that
modifies BIOS settings in the CMOS memory.
This program is available at system startup
time by pressing a key combination such as
Alt+F1 or Ctrl+F2.
BIOS shadow
memory.

A copy of the BIOS in

589

block transfer mode Moving data in discrete groups of bytes, referred to as blocks.
Blue Screen of Death (BSOD) A typical
way of describing the blue screen error condition that occurs when Windows NT fails to
boot properly or quits unexpectedly.
boot The loading of an operating system into
memory, usually from a hard disk, although
occasionally from a floppy disk. This is an
automatic procedure begun when you first turn
on or reset your computer. A set of instructions
contained in ROM begin executing, first running a series of power on self-tests (POSTs) to
check that devices, such as hard disks, are in
working order, then locating and loading the
operating system, and finally passing control of
the computer over to that operating system.

bit Contraction of BInary digiT. A bit is the

basic unit of information in the binary numbering system, representing either 0 (for off) or
1 (for on). Bits can be grouped together to make
up larger storage units, the most common being
the 8-bit byte. A byte can represent all kinds of
information including the letters of the
alphabet, the numbers 0 through 9, and
common punctuation symbols.

bootable disk Any disk capable of loading

and starting the operating system, although
most often used when referring to a floppy
disk. In these days of larger and larger operating systems, it is less common to boot from a
floppy disk. In some cases, all of the files
needed to start the operating system will not fit
on a single floppy disk, which makes it impossible to boot from a floppy.

bit-mapped font A set of characters in a specific style and size, in which each character is
defined by a pattern of dots. The computer
must keep a complete set of bitmaps for every
font you use on your system, and these bitmaps
can consume large amounts of disk space.

Bootstrap Protocol (BOOTP) RFC951 and

RFC 1084 specification for booting diskless
workstations.

bits per second (bps)

See BPS.

blocker A device that will prohibit a signal

from being transmitted.

BPS (bits per second) A measurement of

how much data (how many bits) is being transmitted in one second. Typically used to describe
the speed of asynchronous communications
(modems).

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

590

Glossary

bridge This type of connectivity device operates in the Data Link layer of the OSI model. It
is used to join similar topologies (Ethernet to
Ethernet, Token Ring to Token Ring) and to
divide traffic on network segments. This device
will pass information destined for one particular workstation to that segment, but it will not
pass broadcast traffic.
British Naval Connector (BNC) Barrel connector used to join two segments of coaxial
cable. Also referred to as Bayonet NeillConcelman, named for Paul Neill of Bell Labs
and Carl Concelman.
broadcasting Sending a signal to all entities
that can listen to it. In networking, it refers to
sending a signal to all entities connected to that
network.
brouter In networking, a device that combines the attributes of a bridge and a router. A
brouter can route one or more specific protocols, such as TCP/IP, and bridge all others.
brownout A short period of low voltage,
which is often caused by an unusually heavy
demand for power.
browser A piece of software used to access
the Internet. Common browsers are Netscapes
Navigator and Microsofts Internet Explorer.
bubble-jet printer A type of sprayed ink
printer, this type uses an electric signal that
energizes a heating element, causing ink to
vaporize and get pushed out of the pinhole and
onto the paper.

buffered memory A region of memory

storage that is used to hold data that is waiting
for transfer to or from an I/O device.
bug A logical or programming error in hardware or software that causes a malfunction of
some sort. If the problem is in software, it can
be fixed by changes to the program. If the fault
is in hardware, new circuits must be designed
and constructed. Some bugs are fatal and cause
the program to hang or cause data loss, others
are just annoying, and many are never even
noticed.
bug fix A release of hardware or software that
corrects known bugs but does not contain additional new features. Such releases are usually designated only by an increase in the decimal portion
of the version number; for example, the revision
level may advance from 2 to 2.01 or 2.1, rather
than from 2 to 3.
bus A set of pathways that allow information
and signals to travel between components
inside or outside of a computer.
bus arbitration Negotiation process
between two areas of a computer bus, especially in a PCI Bus.
bus clock A chip on the motherboard that
produces a type of signal (called a clock signal)
that indicates how fast the bus can transmit
information.
bus connector slot A slot made up of several small copper channels that grab the
matching fingers of the expansion circuit
boards. The fingers connect to copper pathways on the motherboard.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

bus controller
computer bus.

Chipset that manages the

bus mastering A technique that allows certain advanced bus architectures to delegate
control of data transfers between the Central
Processing Unit (CPU) and associated peripheral devices to an add-in board.
bus mouse A mouse connected to the computer using an expansion board plugged into an
expansion slot, instead of simply connected to a
serial port as in the case of a serial mouse.
bus slave
master.

I/O devices managed by a bus

bus topology Type of physical topology that

consists of a single cable that runs to every
workstation on the network. Each computer
shares that same data and address path. As
messages pass through the trunk, each workstation checks to see if the message is addressed
for itself. This topology is very difficult to
reconfigure, since reconfiguration requires you
to disconnect and reconnect a portion of the
network (thus bringing the whole network
down).
byte Contraction of BinarY digiT Eight. A
group of 8 bits that, in computer storage terms,
usually holds a single character, such as a
number, letter, or other symbol.
cable access methods Methods by which
stations on a network get permission to
transmit their data.

591

cable select (CSEL) A method for IDE

devices to determine which device is the master
and the C: drive and which device is the slave
and the D: drive.
cable tray

A method of cable management.

cache Pronounced cash. A special area of

memory, managed by a cache controller, that
improves performance by storing the contents
of frequently accessed memory locations and
their addresses. When the processor references
a memory address, the cache checks to see if it
holds that address. If it does, the information is
passed directly to the processor; if not, a
normal memory access takes place instead. A
cache can speed up operations in a computer in
which RAM access is slow compared with its
processor speed, because the cache memory is
always faster than normal RAM.
cache memory Fast SRAM memory used to
store, or cache, frequently used instructions
and data.
capacitive keyboard Keyboard designed with
two sheets of semi-conductive material separated
by a thin sheet of Mylar inside the keyboard. When
a key is pressed, the plunger presses down and a
paddle connected to the plunger presses the two
sheets of semi-conductive material together,
changing the total capacitance of the two sheets.
The controller can tell by the capacitance value
returned which key was pressed.

cable management Providing a method for

managing runs of a large number of cables.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

592

Glossary

capacitive touch screen Type of display

monitor that has two clear plastic coatings over
the screen, separated by air. When the user
presses the screen in a particular spot, the coatings are pressed together and the controller registers a change in the total capacitance of the
two layers. The controller then determines
where the screen was pressed by the capacitance values and sends that information to the
computer in the form of x,y coordinates.
capacitor An electrical component, normally
found in power supplies and timing circuits,
used to store electrical charge.
card services Part of the software support
needed for PCMCIA (PC Card) hardware
devices in a portable computer, controlling the
use of system interrupts, memory, or power
management. When an application wants to
access a PC Card, it always goes through the
card services software and never communicates
directly with the underlying hardware.
carpal tunnel syndrome A form of wrist
injury caused by holding the hands in an awkward position for long periods of time.
carriage motor Stepper motor used to move
the print head back and forth on a dot-matrix
printer.
Carrier Sense Multiple Access/Collision
Detection (CSMA/CD) A network protocol
for managing the situation where two or more
nodes are transmitting at the same time.
cathode-ray tube

See CRT.

CCD (charge-coupled device) A device that

allows light to be converted into electrical
pulses.

CCITT Acronym for Comit Consultatif

Internationale de Tlphonie et de Tlgraphie.
An organization, based in Geneva, that
develops worldwide data communications
standards. CCITT is part of the ITU (International Telecommunications Union). The organization has been renamed ITU-T (ITU
Telecommunications Standardization Sector).
CD-ROM Acronym for compact disc readonly memory. A high-capacity, optical storage
device that uses compact disc technology to
store large amounts of information, up to
650MB (the equivalent of approximately
300,000 pages of text), on a single 4.72" disk.
Central Processing Unit (CPU) The computing and control part of the computer. The
CPU in a mainframe computer may be contained on many printed circuit boards, the CPU
in a mini computer may be contained on several
boards, and the CPU in a PC is contained in a
single extremely powerful microprocessor.
centralized processing A network processing scheme in which all intelligence is
found in one computer and all other computers
send requests to the central computer to be processed. Mainframe networks use centralized
processing.
Centronics parallel interface A standard
36-pin interface in the PC world for the
exchange of information between the PC and a
peripheral, such as a printer, originally developed by the printer manufacturer Centronics,
Inc. The standard defines eight parallel data
lines, plus additional lines for status and control information.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

Certificate Authority (CA) A network service that is the authority that can issue and
manage security credentials and public keys to
be used in message encryption.
CGA Acronym for Color/Graphics Adapter.
CGA is a video adapter that provided lowresolution text and graphics. CGA provided
several different text and graphics modes,
including 40- or 80-column by 25-line, 16color text mode, and graphics modes of 640
horizontal pixels by 200 vertical pixels with
two colors, or 320 horizontal pixels by 200 vertical pixels with four colors. CGA has been
superseded by later video standards, including
EGA, VGA, SuperVGA, and XGA.
channel A path or link through which information is transmitted between connected
devices.
charge-coupled device

See CCD.

charging corona The wire or roller that is

used to put a uniform charge on the EP drum
inside a toner cartridge.
checksum A method of providing information for error detection, usually calculated by
summing a set of values.
checksumming An error checking routine
the runs a mathematical equation against a set
of data and comes up with a result, called a
checksum. The data is then transmitted, and
the receiver then runs the same formula against
the data transmitted and compares the result to
the checksum. If they are the same, the transmission is considered successful.

593

chip creep The slow self-loosening of chips

from their sockets on the system board as a
result of the frequent heating and cooling of the
board (which causes parts of the boardsignificantly, the chip connector slotsto alternately
expand and shrink).
chip puller A tool that is used on older
(pre-386) systems to remove the chips
without damaging them.
cleaning step The step in the EP print process where excess toner is scraped from the EP
drum with a rubber blade.
client A network entity that can request
resources from the network or server.
client computers A computer that requests
resources from a network.
client/server computing An arrangement
on a network that makes use of intelligence on
both the part of the client computer and of the
server service. This type of computing is most
prevalent when working with databases.
client software Software that allows a
device to request resources from a network.
clock doubling Technology that allows a
chip to run at the buss rated speed externally,
but still be able to run the processors internal
clock at twice the speed of the bus. This technology improves computer performance.
clock rate

See clock speed.

clock signal Built-in metronome-like signal

that indicates how fast the components can
operate.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

594

Glossary

clock speed Also known as clock rate. The

internal speed of a computer or processor, normally expressed in MHz. The faster the clock
speed, the faster the computer will perform a
specific operation, assuming the other components in the system, such as disk drives, can
keep up with the increased speed.
clock tripling A type of processor design
where the processor runs at one speed externally and at triple that speed internally.
cluster The smallest unit of hard disk space
that DOS can allocate to a file, consisting of
one or more contiguous sectors. The number of
sectors contained in a cluster depends on the
hard disk type.
cluster servers A group of servers that are
configured so they appear to be one unit providing services to the network.
CMOS Acronym for Complementary Metal
Oxide Semiconductor. An area of nonvolatile
memory that contains settings that determine
how a computer is configured.
CMOS battery A battery used to power
CMOS memory so that the computer wont
loose its settings when powered down.
cold start Turning a computer on and having
it boot up and run.
COMMAND.COM Takes commands issued
by the user through text strings or click actions
and translates them back into calls that can be
understood by the lower layers of DOS. It is the
vital command interpreter for DOS.
community In SNMP, a group of agents
under the control of an SNMP Manager.
Complementary Metal Oxide Semiconductor
See CMOS.

Complex Instruction Set Computing

(CISC) This is a microprocessor implementation
where the processor handles complex sets of
instructions at the same time (as opposed to
Reduced Instruction Set Computing (RISC)).
Component Interface (CI) The interface
between different smaller subsystems in a server.
computer name The name by which a
Microsoft computer is known on the network.
This is a NetBIOS name (up to 15 characters in
length) which must be unique on the network.
In Windows 2000, the computer name is
always the same as the machines host name,
while in Windows 9x the two can be different.
concentrator Also referred to as a hub. Used
in Ethernet networks to connect large numbers
of hosts together.
conditioning step The step in the EP print
process where a uniform charge is applied to
the EP drum by the charging corona or
charging roller.
conductor Any item that permits the flow of
electricity between two entities.
conduit

A pathway for cable.

CONFIG.SYS In DOS and OS/2, a special text

file containing settings that control the way that
the operating system works. CONFIG.SYS must
be located in the root directory of the default boot
disk, normally drive C, and is read by the operating
system only once as the system starts running.
Some application programs and peripheral devices
require you to include special statements in
CONFIG.SYS, while other commands may specify
the number of disk-read buffers or open files on
your system, specify how the disk cache should be
configured, or load any special device drivers your
system may need.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

connectivity device Any device that facilitates connections between network devices.
Some examples include hubs, routers, switches,
and gateways.
Control Program for Microcomputer (CP/M)
A computer operating system that was an
early competitor of Microsofts DOS system.
CP/M was a command-line system developed
by Gary Kildall.
conventional memory The amount of
memory accessible by DOS in PCs using an
Intel processor operating in real mode, normally the first 640K.
cooperative multitasking A form of multitasking in which all running applications must
work together to share system resources.
copy backup A backup of data made by
simply copying the data from one location to
another. For example, you want to make sure
that you have a complete copy of this glossary
file. You would copy it to another location, like
a diskette, for safekeeping.
corona roller Type of transfer corona
assembly that uses a charged roller to apply
charge to the paper.
corona wire Type of transfer corona
assembly. Also, the wire in that assembly that is
charged by the high-voltage supply. It is narrow
in diameter and located in a special notch under
the EP print cartridge.
CPU (Central Processing Unit)
Processing Unit (CPU).

See Central

CPU affinity Also called processor affinity, this

is the process of selecting specific applications or
processes to run on a specific processor.

595

CPU clock Type of clock signal that dictates

how fast the CPU can run.
cross over cable A cable designed to hook
two network cards together without the use of
a hub or a concentrator.
crosstalk Problem related to electromagnetic
fields when two wires carrying electrical signals
run parallel and one of the wires induces a
signal in the second wire. If these wires are carrying data, the extra, unintended signal can
cause errors in the communication. Crosstalk is
especially a problem in unshielded parallel
cables that are longer than 10 feet.
CRT Acronym for cathode-ray tube. A display device used in computer monitors and television sets. A CRT display consists of a glass
vacuum tube that contains one electron gun for
a monochrome display, or three (red, green,
and blue) electron guns for a color display.
Electron beams from these guns sweep rapidly
across the inside of the screen from the upperleft to the lower-right of the screen. The inside
of the screen is coated with thousands of phosphor dots that glow when they are struck by the
electron beam. To stop the image from flickering, the beams sweep at a rate of between 43
and 87 times per second, depending on the
phosphor persistence and the scanning mode
usedinterlaced or non-interlaced. This is
known as the refresh rate and is measured in
Hz. The Video Electronics Standards Association (VESA) recommends a vertical refresh rate
of 72Hz, noninterlaced, at a resolution of 800
by 600 pixels.
Ctrl-Alt-Delete Key sequence to warm boot a
DOS computer or start the shutdown sequence
in a Windows NT/2000 server.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

596

Glossary

Cyclic Redundancy Check (CRC) A method

of checking the validity of computer transmissions.
cylinder A hard disk consists of two or more
platters, each with two sides. Each side is further divided into concentric circles known as
tracks, and all the tracks at the same concentric
position on a disk are known collectively as a
cylinder.
daisy-chaining Pattern of cabling where the
cables run from the first device to the second,
second to the third, and so on. If the devices
have both an in and an out, the in of the
first device of each pair is connected to the out
of the second device of each pair.
daisy-wheel printer An impact printer that
uses a plastic or metal print mechanism with a
different character on the end of each spoke of
the wheel. As the print mechanism rotates to
the correct letter, a small hammer strikes the
character against the ribbon, transferring the
image onto the paper.
DAT

See digital audio tape (DAT).

data bits In asynchronous transmissions, the

bits that actually comprise the data; usually 7
or 8 data bits make up the data word.
data bus Bus used to send and receive data to
the microprocessor.
data compression Any method of encoding
data so that it occupies less space than in its
original form.

has remained unbroken despite years of use; it

completely randomizes the information so that
it is impossible to determine the encryption key
even if some of the original text is known.
Data Link layer The second of seven layers
of the International Standards Organizations
Open Systems Interconnection (ISO/OSI)
model for computer-to-computer communications. The Data Link layer validates the integrity of the flow of data from one node to
another by synchronizing blocks of data and by
controlling the flow of data.
data set ready

See DSR.

data striping A combination of the same size

disk partitions that reside on separate disks.
These partitions are joined to form a single
volume by forming a virtual stripe across the
disks that the network operating system recognizes as a single drive.
data terminal equipment
data terminal ready

See DTE.

See DTR.

data transfer rate The speed at which a disk

drive can transfer information from the drive to
the processor, usually measured in megabits or
megabytes per second.
data warehousing Storing data in several
locations, on several servers.
database server A server dedicated to
storing, retrieving, and manipulating large
amounts of data quickly.

data encoding scheme (DES) The method

used by a disk controller to store digital information onto a hard disk or floppy disk. DES

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

daughter board A printed circuit board that

attaches to another board to provide additional
functions.
DB connector Any of several types of cable
connectors used for parallel or serial cables.
The number following the letters DB (for data
bus) indicates the number of pins that the connector usually has.
de facto Latin translation for by fact. Any
standard that is a standard because everyone is
using it.
de jure Latin translation for by law. Any
standard that is a standard because a standards
body decided it should be so.
debouncing A keyboard feature that eliminates unintended triggering of keystrokes. It
works by having the keyboard controller constantly scan the keyboard for keystrokes. Only
keystrokes that are pressed for more than two
scans are considered keystrokes. This prevents
spurious electronic signals from generating input.
decimal The base-10 numbering system that
uses the familiar numbers 09.
dedicated application server A server, usually a Web server, dedicated to providing lineof-business applications.
dedicated server The server that is assigned
to perform a specific application or service.
default gateway If a user needs to communicate by TCP/IP with a computer that is not on
their subnet (the local network segment), the
computer needs to use a gateway to access this
remote network. The default gateway is simply
the path that is taken by all outgoing traffic
unless another path is specified.

597

defragmentation The process of reorganizing and rewriting files so that they occupy
one large continuous area on your hard disk
rather than several smaller areas.
DES

See data encoding scheme (DES).

desktop Contains the visible elements of

Windows and defines the limits of graphic
environment.
Desktop Control Panel Windows panel that
is used to configure the system so it is more
easily usable. This control panel contains the
settings for the background color and pattern
as well as screen saver settings.
Desktop Management Task Force
(DMTF) Task Force charged with developing
the protocols and procedures necessary for efficient remote management of computer desktops.
developing roller The roller inside a toner
cartridge that presents a uniform line of toner
to help apply the toner to the image written on
the EP drum.
developing step The step in the EP print
process where the image written on the EP
drum by the laser is developed, that is, it has
toner stuck to it.
device driver A small program that allows a
computer to communicate with and control a
device.
Device Manager A utility in Windows 9x
and Windows 2000 that allows the user to view
and modify hardware settings. Device drivers
can be installed or upgraded, and problems
with devices can be found and dealt with here.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

598

Glossary

DEVICE= Command found in the DOS

CONFIG.SYS that tells DOS which driver to
find and load into memory at boot time.
DEVICEHIGH= Command that is used to
load the device drivers into upper memory
blocks, thereby freeing up space in conventional memory.
DHCP Relay Agent A computer on a subnet
that does not have its own DHCP server that
listens for DHCP requests. If it hears a
DHCP request, the relay agent forwards the
request in a directed packet to the designated
DHCP server on another subnet.
diagnostic program A program that tests
computer hardware and peripherals for correct
operation. In the PC, some faults are easy to
find, and these are known as hard faults; the
diagnostic program will diagnose them correctly every time. Others, such as memory
faults, can be difficult to find; these are called
soft faults because they do not occur every
time the memory location is tested, but only
under very specific circumstances.
differential backup Backs up files that have
changed since the last full backup.
digital audio tape (DAT) A method of
recording information in digital form on a
small audio tape cassette. Many gigabytes of
information can be recorded on a cassette, and
so a DAT can be used as a backup medium.
Like all tape devices, however, DATs are relatively slow.
digital signal A signal that consists of discrete values. These values do not change over
time; in effect, they change instantly from one
value to another.

DIMM (Dual Inline Memory Module)

Memory module that is similar to a SIMM
(Single Inline Memory Module), except that a
DIMM is double-sided. There are memory
chips on both sides of the memory module.
DIN-n Circular type of connector used with
computers. (The n represents the number of
connectors.)
DIP (Dual Inline Package) A standard
housing constructed of hard plastic commonly
used to hold an integrated circuit. The circuits
leads are connected to two parallel rows of pins
designed to fit snugly into a socket; these pins
may also be soldered directly to a printedcircuit board. If you try to install or remove
dual inline packages, be careful not to bend or
damage their pins.
DIP switch A small switch used to select the
operating mode of a device, mounted as a Dual
Inline Package. DIP switches can be either
sliding switches or rocker switches and are
often grouped together for convenience. They
are used on printed circuit boards, dot-matrix
printers, modems, and other peripherals.
direct memory access
memory access).

See DMA (direct

Direct Memory Execute (DIME) DIME is a

feature of AGP video. With DIME, AGP graphics
chips have the capability to access main memory
directly for the complex operations.
directory Directories are used to organize
files on the hard drive. Another name for a
directory is a folder. Directories created inside
or below others are called subfolders or
subdirectories.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

599

Direct Rambus A memory bus that transfers

data at 800MHz over a 16-bit memory bus.
Direct Rambus memory models (often called
RIMMs), like DDR SDRAM, can transfer data
on both the rising and falling edges of a clock
cycle.

disk drive A peripheral storage device that

reads and writes to magnetic or optical disks.
When more than one disk drive is installed on a
computer, the operating system assigns each
drive a unique namefor example A and C in
DOS, Windows, and OS/2.

direct-solder method A method of

attaching chips to the motherboard where the
chip is soldered directly to the motherboard.

disk duplexing In networking, a fault-tolerant technique that writes the same information simultaneously onto two different hard
disks. Disk duplexing is offered by most of the
major network operating systems and is
designed to protect the system against a single
disk failure; it is not designed to protect against
multiple disk failures and is no substitute for a
well-planned series of disk backups.

disaster recovery The process of planning

for and executing procedures to recover the
computer network after a disaster.
disk cache An area of computer memory
where data is temporarily stored on its way to
or from a disk. A disk cache mediates between
the application and the hard disk, and when an
application asks for information from the hard
disk, the cache program first checks to see if
that data is already in the cache memory. If it is,
the disk cache program loads the information
from the cache memory rather than from the
hard disk. If the information is not in memory,
the cache program reads the data from the disk,
copies it into the cache memory for future reference, and then passes the data to the
requesting application.
disk-cacheing program A program that
reads the most commonly accessed data from
disk and keeps it in memory for faster access.
disk controller The electronic circuitry that
controls and manages the operation of floppy
or hard disks installed in the computer. A single
disk controller may manage more than one
hard disk; many disk controllers also manage
floppy disks and compatible tape drives.

diskette An easily removable and portable

floppy disk that is 3.5" in diameter and
enclosed in a durable plastic case that has a
metal shutter over the media access window.
diskless workstation A networked computer that does not have any local disk storage
capability.
disk mirroring In networking, a faulttolerant technique that writes the same information simultaneously onto two different hard
disks, using the same disk controller. In the
event of one disk failing, information from the
other can be used to continue operations. Disk
mirroring is offered by most of the major network operating systems and is designed to protect the system against a single disk failure; it is
not designed to protect against multiple disk
failures and is no substitute for a well-planned
series of disk backups.
Disk Operating System

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

See DOS.

600

Glossary

distributed application server A group of

servers spread out over an enterprise to provide
application support.
distributed processing A computer system
in which processing is performed by several
separate computers linked by a communications network. The term often refers to any
computer system supported by a network, but
more properly refers to a system in which each
computer is chosen to handle a specific workload and the network supports the system as a
whole.
DIX Ethernet The original name for the
Ethernet network technology. Named after the
original developer companies, Digital, Intel,
and Xerox.
DMA (direct memory access) A method of
transferring information directly from a massstorage device such as a hard disk or from an
adapter card into memory (or vice versa),
without the information passing through the
processor.
DMA channels Dedicated circuit pathways
on the motherboard that make DMA possible.
docking station A hardware system into
which a portable computer fits so that it can be
used as a full-fledged desktop computer.
Docking stations vary from simple port replicators (that allow you access to parallel and serial
ports and a mouse) to complete systems (that
give you access to network connections, CDROMs, even a tape backup system or PCMCIA
ports).

domain 1. The security structure for Windows

NT Server and Windows 2000 Active Directory.
2. The namespace structure of TCP/IPs DNS
structure.
Domain Name System (DNS) DNS
allows TCP/IP-capable users anywhere in the
world to find resources in other companies or
countries by using their domain name. Each
domain is an independent namespace for a particular organization, and DNS servers manage
requests for information about the IP addresses
of particular DNS entries. DNS is used to
manage all names on the Internet.
dongle A special cable that provides a connector to a circuit board that doesnt have one.
For example, a motherboard may use a dongle
to provide a serial port when there is a ribbon
cable connector for the dongle on the motherboard, but there is no serial port.
dongle connection A connector on a motherboard where a dongle will connect.
DOS 1. Acronym for Disk Operating System,
an operating system originally developed by
Microsoft for the IBM PC. DOS exists in two
very similar versions; MS-DOS, developed and
marketed by Microsoft for use with IBMcompatible computers, and PC-DOS, supported
and sold by IBM for use only on computers
manufactured by IBM. 2. A DOS CONFIG.SYS
command that loads the operating system into
conventional memory, extended memory, or
into upper memory blocks on computers using
the Intel 80386 or later processor. To use this
command, you must have previously loaded the
HIMEM.SYS device driver with the DEVICE
command in CONFIG.SYS.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

DOS Environment Variables Variables that

specify global things like the path that DOS
searches to find executables.
DOS extender A small program that extends
the range of DOS memory. For example,
HIMEM.SYS allows DOS access to the
memory ranges about 1024K.
DOS prompt A visual confirmation that
DOS is ready to receive input from the keyboard. The default prompt includes the current
drive letter followed by a right angle bracket
(for example, C>). You can create your own
custom prompt with the PROMPT command.
DOS shell An early graphic user interface for
DOS that allowed users to manage files and run
programs through a simple text interface and
even use a mouse. It was soon replaced by Windows.
dot-matrix printer An impact printer that
uses columns of small pins and an inked ribbon
to create the tiny pattern of dots that form the
characters. Dot-matrix printers are available in
9-, 18-, or 24-pin configurations.
dot pitch In a monitor, the vertical distance
between the centers of like-colored phosphors
on the screen of a color monitor, measured in
millimeters (mm).
dots per inch (dpi) A measure of resolution
expressed by the number of dots that a device
can print or display in one inch.
Double Data Rate SDRAM (DDR SDRAM)
One of the newer technologies to be instituted for Random Access Memory, which
increases the amount of data that can be processed from the system bus.

601

double-density disk A floppy disk with a

storage capacity of 360KB.
Down command Command issued to halt a
Novell NetWare server.
DRAM

See dynamic RAM (DRAM).

drawing tablet Pointing device that includes

a pencil-like device (called a stylus) for drawing
on its flat rubber-coated sheet of plastic.
drive bay An opening in the system unit into
which you can install a floppy disk drive, hard
disk drive, or tape drive.
drive geometry Term used to describe the
number of cylinders, read/write heads, and sectors in a hard disk.
drive hole Hole in a floppy disk that allows
the motor in the disk drive to spin the disk. Also
known as the hub hole.
drive letter In DOS, Windows, and OS/2,
the drive letter is a designation used to specify a
particular hard or floppy disk. For example, the
first floppy disk is usually referred to as drive A,
and the first hard disk as drive C.
driver

See device driver.

driver signing In order to prevent viruses

and poorly written drivers from damaging your
system, Windows 2000 uses a process called
driver signing that allows companies to digitally sign their device software, and it also
allows administrators to block the installation
of unsigned drivers.
driver software
D-Shell

Copyright 2001 SYBEX, Inc., Alameda, CA

See device driver.

See DB connector.

www.sybex.com

602

Glossary

DSR Abbreviation for data set ready. A hardware signal defined by the RS-232-C standard
to indicate that the device is ready.
D-Sub

See DB connector.

DTE Abbreviation for data terminal equipment. In communications, any device, such as a
terminal or a computer, connected to a communications channel or public network.
DTR Abbreviation for data terminal ready. A
hardware signal defined by the RS-232-C standard to indicate that the computer is ready to
accept a transmission.
dual-booting If a single machine must be
used for many tasks, it may be necessary for it
to have multiple operating systems installed
simultaneously. To do this a boot manager presents the user with a choice of which operating
system to use at startup. To use a different OS
the user would have to shut down the system,
restart it, and select the other OS.
dual processing (DP) Having two processors working in conjunction with each other on
a motherboard.
Dual Inline Memory Module See DIMM
(Dual Inline Memory Module).
Dual Inline Package
Package).

See DIP (Dual Inline

dumb terminal A combination of keyboard

and screen that has no local computing power,
used to input information to a large, remote
computer, often a minicomputer or a mainframe. This remote computer provides all the
processing power for the system.

duplex In asynchronous transmissions, the

ability to transmit and receive on the same
channel at the same time; also referred to as full
duplex. Half duplex channels can transmit only
or receive only. Most dial-up services available
to PC users take advantage of full duplex capabilities, but if you cannot see what you are
typing, switch to half duplex. If you are using
half duplex and you can see two of every character you type, change to full duplex.
duplex printing Printing a document on
both sides of the page so that the appropriate
pages face each other when the document is
bound.
Dynamic DNS (DDNS) A process where
Domain Name Service entries are made automatically by DHCP servers. This is part of the
technology of IP version 6.
dynamic electricity

See electricity.

Dynamic Host Configuration Protocol

(DHCP) DHCP manages the automatic
assignment of TCP/IP addressing information
(such as the IP address, subnet mask, default
gateway and DNS server). This can save a great
deal of time when configuring and maintaining
a TCP/IP network.
Dynamic Link Library (DLL) files Windows
component files that contain small pieces of
executable code that are shared between multiple Windows programs. They are used to
eliminate redundant programming in certain
Windows applications. DLLs are used extensively in Microsoft Windows, OS/2, and in
Windows NT. DLLs may have filename extensions of .dll, .drv, or .fon.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

dynamic RAM (DRAM) A common type of

computer memory that uses capacitors and
transistors storing electrical charges to represent memory states. These capacitors lose their
electrical charge, and so they need to be
refreshed every millisecond, during which time
they cannot be read by the processor. DRAM
chips are small, simple, cheap, easy to make,
and hold approximately four times as much
information as a static RAM (SRAM) chip of
similar complexity. However, they are slower
than static RAM. Processors operating at clock
speeds of 25MHz or more need DRAM with
access times of faster than 80 nanoseconds (80
billionths of a second), while SRAM chips can
be read in as little as 15 to 30 nanoseconds.
Each operating system contains a standard
set of device drivers for the keyboard, the monitor, and so on, but if you add specialized
peripherals (such as a CD-ROM disk drive) or
a network interface card, you will probably
have to add the appropriate device driver so
that the operating system knows how to
manage the device. In DOS, device drivers are
loaded by the DEVICE or DEVICEHIGH commands in CONFIG.SYS.
edge connector A form of connector consisting of a row of etched contacts along the
edge of a printed circuit board that is inserted
into an expansion slot in the computer.
EDO (Extended Data Out) RAM A type of
DRAM that increases memory performance by
eliminating wait states.
EEPROM Acronym for Electrically Erasable
Programmable Read-Only Memory. A
memory chip that maintains its contents
without electrical power, and whose contents
can be erased and reprogrammed either within

603

the computer or from an external source.

EEPROMs are used where the application
requires stable storage without power but may
have to be reprogrammed.
EGA Acronym for Enhanced Graphics
Adapter. A video adapter standard that provides medium-resolution text and graphics.
EGA can display 16 colors at the same time
from a choice of 64, with a horizontal resolution of 640 pixels and a vertical resolution of
350 pixels. EGA has been superseded by VGA
and SVGA.
EISA Acronym for Extended Industry Standard Architecture. A PC bus standard that
extends the traditional AT-bus to 32 bits and
allows more than one processor to share the
bus. EISA has a 32-bit data path and, at a bus
speed of 8MHz, can achieve a maximum
throughput of 33 megabytes per second.
EISA Configuration Utility (EISA Config)
The utility used to configure an EISA bus
expansion card.
Electrically Erasable Programmable ReadOnly Memory See EEPROM.
electricity The flow of free electrons from
one molecule of substance to another. This
flow of electrons is used to do work.
electromagnetic drawing tablets Type of
drawing tablet that has grids of wires underneath the rubberized surface. The stylus contains a small sensor that is sensitive to
electromagnetic fields. At timed intervals, an
electromagnetic pulse is sent across the grid.
The sensor in the stylus picks up these pulses.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

604

Glossary

electromagnetic interference (EMI) Any

electromagnetic radiation released by an electronic device that disrupts the operation or performance of any other device.
electron gun The component of a monitor
that fires electrons at the back of the phosphorcoated screen.
Electronics Industry Association (EIA) An
association based in Washington DC that is
designed to serve the needs of various manufacturers of electronic components. This association sets standards for these components.
electrostatic discharge (ESD) When two
objects of dissimilar charge come in contact
with one another, they will exchange electrons
in order to standardize the electrostatic charge
between the two objects. This exchange, or discharge, can sometimes be seen as a spark or arc
of electricity. Even when it cannot be seen it is
damaging to electronic components.
e-mail Electronic mail is generally sent across
the Internet using protocols named SMTP (for
sending) and POP3 (for receiving).
EMI See electromagnetic interference
(EMI).
EMM386.EXE Reserved memory manager
that emulates Expanded Memory in the
Extended Memory area (XMS) and provides
DOS with the ability to utilize upper memory
blocks to load programs and device drivers.
encoding Process by which binary information is changed into flux transition patterns on
a disk surface.

Encrypting File System (EFS) File system

used in Windows 2000 to encrypt data stored
on local and network drives.
Enhanced Graphics Adapter

See EGA.

Enhanced IDE (EIDE) An extension of the

base IDE (ATA) standard for disk drives and
other devices. EIDE allows for faster data
transfer and more devices on the IDE chain.
enhanced keyboard A 101- or 102-key keyboard introduced by IBM that has become the
accepted standard for PC keyboard layout.
Unlike earlier keyboards, it has 12 function
keys across the top, rather than 10 function
keys in a block on the left side, has extra Ctrl
and Alt keys, and has a set of cursor control
keys between the main keyboard and the
numeric keypad.
Enhanced Small Device Interface (ESDI) A
popular hard-disk, floppy-disk, and tape-drive
interface standard, capable of a data transfer
rate of 10 to 20 megabits per second. ESDI is
most often used with large hard disks.
environment variables These are used to set
certain system-wide parameters that can then
be used by applications running on the system.
For instance, a systems temporary directory
can be set to a specific location using an environment variable.
EP drum Device that is coated with a photosensitive material that can hold a static charge
when not exposed to light. The drum contains
a cleaning blade that continuously scrapes the
used toner off the photosensitive drum to keep
it clean.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

605

EP print process Six-step process an EP laser

printer uses to form images on paper. In order,
the steps are charging, exposing, developing,
transferring, fusing, and cleaning.

even parity A technique that counts the

number of 1s in a binary number and, if the
number of 1s total is not an even number, adds
a digit to make it even. (See also parity).

EP printer (electrophotographic printer)

Printer that uses high voltage, a laser, and a
black carbon toner to form an image on a page.

exit roller Found on laser and page printers,

the mechanism that guides the paper out of the
printer into the paper-receiving tray.

Eraseable Progammable Read Only

Memory (EPROM) Acronym for erasable
programmable read-only memory. A memory
chip that maintains its contents without electrical power, and whose contents can be erased
and reprogrammed by removing a protective
cover and exposing the chip to ultraviolet light.

expanded memory page frame

frame.

ergonomics Standards that define the positioning and use of the body to promote a
healthy work environment.
error correction code (ECC) A method that
allows for the detection and correction of
errors that may occur during data transmission. Error correction codes are categorized by
the numbers of errors that can be corrected.
ESD

See electrostatic discharge (ESD).

ESD mat Preventive measure to guard

against the effects of ESD. The excess charge is
drained away from any item that comes in contact with it.
Ethernet A network technology based on the
IEEE 802.3 CSMA/CD standard. The original
Ethernet implementation specified 10MBps,
baseband signaling, coaxial cable, and CSMA/
CD media access.

See page

expanded memory specification (EMS)

The original version of the Lotus-IntelMicrosoft Expanded Memory Specification
(LIM EMS) that lets DOS applications use
more than 640KB of memory space.
expansion bus An extension of the main
computer bus that includes expansion slots for
use by compatible adapters, such as memory
boards, video adapters, hard disk controllers,
and SCSI interface cards.
expansion card A device that can be
installed into a computers expansion bus.
expansion slot One of the connectors on the
expansion bus that gives an adapter access to
the system bus. You can add as many additional adapters as there are expansion slots
inside your computer.
Extended Data Out (EDO) A type of
random access memory that allows data to stay
in memory for longer periods of time.
extended DOS partition A further optional
division of a hard disk, after the primary DOS
partition, that functions as one or more additional logical drives. A logical drive is simply an
area of a larger disk that acts as though it were
a separate disk with its own drive letter.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

606

Glossary

Extended Graphics Array

See XGA.

Extended Industry Standard Architecture

See EISA.
extended memory manager A device
driver that supports the software portion of the
extended memory specification in an IBMcompatible computer.
Extended Memory System (XMS)
Memory above 1,024KB that is used by
Windows and Windows-based programs. This
type of memory cannot be accessed unless the
HIMEM.SYS memory manager is loaded in the
DOS CONFIG.SYS with a line like
DEVICE=HIMEM.SYS.
extended partition If all of the space on a
drive is not used in the creation of the drives
primary partition, a second partition can be
created out of the remaining space. Called the
extended partition, this second partition can
hold one or more logical drives.
Exterior Gateway Protocol (EGP) A
routing protocol used for providing information on the availability of routers and gateways
that interconnect networks.
external bus An external component connected through expansion cards and slots
allows the processor to talk to other devices.
This component allows the CPU to talk to the
other devices in the computer and vice versa.
external cache memory Separate expansion board that installs in a special processordirect bus that contains cache memory.

external commands Commands that are not

contained within COMMAND.COM. They are
represented by a .COM or .EXE extension.
external hard disk A hard disk packaged in
its own case with cables and an independent
power supply rather than a disk drive housed
inside and integrated with the computers
system unit.
external modem A stand-alone modem,
separate from the computer and connected by a
serial cable. LEDs on the front of the chassis
indicate the current modem status and can be
useful in troubleshooting communications
problems. An external modem is a good buy if
you want to use a modem with different computers at different times or with different types
of computer.
F_port A Fibre Channel Switch port capable
of supporting one N_Port device.
FL_port A Fibre Channel Switch port capable
of supporting NL_Port devices, which implement the Public Loop protocols.
Fabric When discussing Fibre Channel
devices, the Fabric is the infrastructure of the
way the devices are connected together.
Fabric Switched When discussing Fibre
Channel devices, the Fabric is the infrastructure
of the way the devices are connected together.
In this case, the devices would be joined by
switches.
Failure Alert A method for providing notification if a failure or fail over has occurred.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

FAQ Acronym for Frequently Asked Question. A document that lists some of the more
commonly asked questions about a product or
component. When researching a problem, the
FAQ is usually the best place to start.
Fast-ATA Fast-ATA is a term coined by
Seagate Technologies that refers to EIDE. See
EIDE.
Fast Ethernet Ethernet specifications that
match the 100Base-X specifications.
fast paged mode (FPM) Dynamic Random
Access Memory that accesses memory in
sequential order with a reduced cycle time. This
provides for faster throughput.
Fast SCSI A SCSI-2 Interface specification
that transfers data at a rate of 10Mbps. Data is
transferred 8 bits at a time. Fast SCSI uses a 50pin adapter.
Fast-Wide SCSI A SCSI-2 interface specification that transfers data at a rate of up to
20Mbps. Data is transferred 16 bits at a time.
The Fast-Wide SCSI adapter is a 68-pin
adapter.
FAT

See file allocation table (FAT).

fault tolerance Fault tolerance is a computer

system or computer subsystem that is designed
provide a backup for a failed component or service. If a device or service fails, another replaces
it immediately without a loss of service. Fault
tolerance can be provided with either hardware
or software or in a combination of the two.

607

fax server A server process that provides for

the transmission, reception, and delivery of
faxes. This is usually integrated with an e-mail
system.
FDDI See fiber distributed data interface
(FDDI).
FDISK.EXE The DOS utility that is used to
partition hard disks for use with DOS.
feed roller The rubber roller in a laser printer
that feeds the paper into the printer.
fiber distributed data interface (FDDI) A
specification for fiber-optic networks transmitting at a speed of up to 100 megabits per second
over a dual, counter-rotating, Token Ring
topology. FDDI is suited to systems that require
the transfer of very large amounts of information, such as medical imaging, 3D seismic processing, oil reservoir simulation, and fullmotion video.
fiber optic cable A transmission technology
that sends pulses of light along specially manufactured optical fibers. Each fiber consists of a
core, thinner than a human hair, surrounded by
a sheath with a much lower refractive index.
Light signals introduced at one end of the cable
are conducted along the cable as the signals are
reflected from the sheath.
fiber optic inter-repeater link (FOIRL) A
repeater-to-repeater link using fiber optic cable.

fax modem An adapter that fits into a PC

expansion slot and provides many of the capabilities of a full-sized fax machine, but at a fraction of the cost.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

608

Glossary

Fibre Channel Fibre Channel is designed to

be a highly reliable interconnect technology
that can operate at gigabit speeds and can allow
concurrent communications among workstations, mainframes, servers, data storage systems, and other peripherals using SCSI and IP
protocols. Fibre Channel also provides interconnect systems for multiple topologies that
can scale to a total system bandwidth on the
order of a terabit per second.

This unreadable file is referred to as corrupt

and it must be either repaired or replaced.
file locking A feature of many network operating systems that prevents more than one
person from updating a file at the same time by
locking the file.

field replacement unit See FRU (field

replacement unit).

File Manager Windows utility that allows

the user to accomplish a number of important
file-related tasks from a single interface. This is
a Windows 3.x feature only; Window 9x uses
Explorer.

file allocation table (FAT) A table maintained by DOS or OS/2 that lists all the clusters
available on a disk. The FAT includes the location of each cluster, as well as whether it is in
use, available for use, or damaged in some way
and therefore unavailable. FAT also keeps
track of which pieces belong to which file.

file server A networked computer used to

store files for access by other client computers
on the network. On larger networks, the file
server may run a special network operating
system; on smaller installations, the file server
may run a PC operating system supplemented
by peer-to-peer networking software.

file and print server A server that is

designed to provided file services, like the
storage and retrieval of data and applications,
as well as access to network print devices. See
File Server and Print Server.

file sharing In networking, the sharing of

files via the network file server. Shared files can
be read, reviewed, and updated by more than
one individual. Access to the file or files is often
regulated by password protection, account or
security clearance, or file locking, to prevent
simultaneous changes from being made by
more than one person at a time.

file compression program An application

program that shrinks program or data files, so
that they occupy less disk space. The file must
then be extracted or decompressed before you
can use it. Many of the most popular file compression programs are shareware, like WinZIP,
PKZIP, LHA, and StuffIt for the Macintosh,
although utility packages like PC Tools from
Central Point Software also contain file compression programs.
file corruption Occasionally an improper
shutdown, a virus, or a random problem will
cause a files information to become unreadable.

File Transfer Protocol (FTP) FTP is used to

transfer large files across the Internet or any
TCP/IP network. Special servers, called FTP
servers, store information and then transfer it
back to FTP clients as needed. FTP servers can
also be secured with a username and password
to prevent unauthorized downloading
(retrieval of a file from the server) or uploading
(placing of a file on the server).

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

firewall A firewall is a security tool that is

designed to protect a corporate network from
external threats. Firewalls are usually a combination of hardware and software. They also
prevent computers on the corporate network
from accessing the Internet directly.
FireWire

See IEEE-1394.

firmware Any software stored in a form of

read-only memoryROM, EPROM, or
EEPROMthat maintains its contents when
power is removed.
fixed disk A disk drive that contain several
disks (also known as platters) stacked together
and mounted through their centers on a small
rod. The disks rotate as read/write heads float
above the disks that make, modify, or sense
changes in the magnetic positions of the coatings on the disk.
fixed resistor Type of resistor that is used to
reduce the current by a certain amount. Fixed
resistors are color coded to identify their resistance values and tolerance bands.
flash memory A special form of non-volatile
EEPROM that can be erased at signal levels
normally found inside the PC, so that you can
reprogram the contents with whatever you like
without pulling the chips out of your computer.
Also, once flash memory has been programmed, you can remove the expansion board
it is mounted on and plug it into another computer if you wish.
flatbed scanner An optical device used to
digitize a whole page or a large image.

609

flat-panel display In laptop and notebook

computers, a very narrow display that uses one
of several technologies, such as electroluminescence, LCD, or thin film transistors.
floating-point calculation A calculation of
numbers whose decimal point is not fixed but
moves or floats to provide the best degree of
accuracy. Floating-point calculations can be
implemented in software, or they can be performed much faster by a separate floatingpoint processor.
floating-point processor A special-purpose,
secondary processor designed to perform
floating-point calculations much faster than the
main processor.
floppy disk A flat, round, magnetically
coated plastic disk enclosed in a protective
jacket. Data is written onto the floppy disk by
the disk drives read/write heads as the disk
rotates inside the jacket. It can be used to distribute commercial software, to transfer programs from one computer to another, or to
back up files from a hard disk. Floppy disks in
personal computing are of two physical sizes,
5.25" or 3.5", and a variety of storage capacities. The 5.25" floppy disk has a stiff plastic
external cover, while the 3.5" floppy disk is
enclosed in a hard plastic case. IBMcompatibles use 5.25" and 3.5" disks, and the
Macintosh uses 3.5" disks.
floppy disk controller The circuit board that
is installed in a computer to translate signals
from the CPU into signals that the floppy disk
drive can understand. Often it is integrated into
the same circuit board that houses the hard disk
controller; it can, however, be integrated into
the motherboard in the PC.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

610

Glossary

floppy disk drive A device used to read and

write data to and from a floppy disk. Floppy
disk drives may be full-height drives, but more
commonly these days they are half-height
drives.
floppy drive cable A cable that connects the
floppy drive(s) to the floppy drive controller.
The cable is a 34-wire ribbon cable that usually
has three connectors.
floppy drive interfaces A connector on a
motherboard used to connect floppy drives to
the motherboard.
floptical disk A removable optical disk with
a recording capacity of between 20 and 25
megabytes.
flow control A method for limiting the
amount of data that can move between two
computer components at any given time.
flux transition Presence or absence of a magnetic field in a particle of the coating on the
disk. As the disk passes over an area the electromagnet is energized to cause the material to
be magnetized in a small area.
footprint The amount of desktop or floor
space occupied by a computer or display terminal. By extension, also refers to the size of
software items such as applications or operating systems.
Force Perfect Termination (FPT) Force Perfect Termination is a single-ended termination
method that uses diode switching and biasing
to make up for any impedance mismatches that
exist between the SCSI cabling and the peripheral device.

FORMAT.COM External DOS command

that prepares the partition to store information
using the FAT system as required by DOS and
Windows 9x.
formatter board Type of circuit board that
takes the information the printer receives from
the computer and turns it into commands for
the various components in the printer.
formatting 1. To apply the page-layout commands and font specifications to a document
and produce the final printed output.
2. The process of initializing a new, blank
floppy disk or hard disk so that it can be used to
store information.
form factors Physical characteristics and
dimensions of drive styles.
form feed (FF) A printer command that
advances the paper in the printer to the top of
the next page by pressing the FF button on the
printer.
fragmentation A disk storage problem that
exists after several smaller files have been
deleted from a hard disk. The deletion of files
leaves the disk with areas of free disk space
scattered throughout the disk. The fact that
these areas of disk space are located so far apart
on the disk causes slower performance because
the disk read/write heads have to move all
around the disks surface to find the pieces of
one file.
Frame Check Sequence (FCS) The FCS is
part of a frame that is transmitted across an
FDDI network. The FCS is used by a receiving
station to verify that the frame traversed the
network without incurring any bit errors.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

free memory
rently in use.

An area of memory not cur-

frequencies The measure of how often a

periodic event occurs. In electronics, this could
be a signal going through a complete cycle.
Frequently Asked Question

See FAQ.

friction feed A paper-feed mechanism that

uses pinch rollers to move the paper through a
printer, one page at a time.
Front Side Bus (FSB) Also known as the
system bus. See system bus.
FRU (field replacement unit) The individual parts or whole assemblies that can be
replaced to repair a computer.
Full AT A type of motherboard form factor
where the motherboard is the same size as the
original IBM AT computers motherboard.
Full backup A backup strategy where everything on the drive has been backed up,
including the operating system, applications,
and all data.
full duplex
tions.

See full duplex communica-

full duplex mode

nications.

See full duplex commu-

full duplex communications Communications where both entities can send and receive
simultaneously.
Full Hot Swap This term is used in PCI Hot
Swap devices. It means that the system bus is
fully Hot Swap capable, where you can remove
and reinstall a component without taking the
entire system out of service.

611

full zone transfer Used in Domain Name

Service (DNS) to transfer all the of data in one
database to another DNS server somewhere
else on the network.
function keys The set of programmable keys
on the keyboard that can perform special tasks
assigned by the current application program.
fuser Device on an EP Printer that uses two
rollers to heat the toner particles and melt them
to the paper. The fuser is made up of a halogen
heating lamp, a Teflon-coated aluminum fusing
roller, and a rubberized pressure roller. The
lamp heats the aluminum roller. As the paper
passes between the two rollers, the rubber
roller presses the paper against the heated
roller. This causes the toner to melt and become
a permanent image on the paper.
fusing assembly

See fuser.

fusing step The step in the EP process where

the toner image on the paper is fused to the
paper using heat and pressure. The heat melts
the toner and the pressure helps fuse the image
permanently to the paper.
game port A DB-15 connector used to connect game devices (like joysticks) to a computer.
gateway In networking, a shared connection
between a local area network and a larger
system, such as a mainframe computer or a
large packet-switching network. Usually
slower than a bridge or router, a gateway typically has its own processor and memory and
can perform protocol conversions. Protocol
conversion allows a gateway to connect two
dissimilar networks; data is converted and
reformatted before it is forwarded to the new
network.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

612

Glossary

GDI.EXE Windows core component that is

responsible for drawing icons and windows in
Windows 3.x.

group icons A type of Windows icon that

groups Windows program icons together in the
Program Manager.

General Protection Fault (GPF) A Windows error that typically occurs when a Windows program tries to access memory currently
in use by another program.

Group Policies A method of applying security or auditing functions to a group of users or

computers in a Windows 2000 network.

gigabit Describes 1,000 million bits. In data

storage, it is 1024 X 1,048,576.
Gigabit Ethernet IEEE 802.3z, which
includes support for transmission at speeds of
1000Mbps.
gigabyte One billion bytes; however, bytes
are most often counted in powers of 2, and so a
gigabyte becomes 2 to the 30th power, or
1,073,741,824 bytes.
GPF

See General Protection Fault (GPF).

graphical user interface (GUI) A graphicsbased user interface that allows users to select
files, programs, or commands by pointing to pictorial representations on the screen rather than by
typing long, complex commands from a command prompt. Application programs execute in
windows, using a consistent set of pull-down
menus, dialog boxes, and other graphical elements such as scroll bars and icons.
graphics accelerator board A specialized
expansion board containing a graphics coprocessor as well as all the other circuitry found on
a video adapter.

GUI

See graphical user interface (GUI).

half duplex

See half duplex communications.

half duplex mode

cations.

See half duplex communi-

half duplex communications Communications that occur when only one entity can transmit
or receive at any one instant.
half-height drive A space-saving drive bay
that is half the height of the 3" drive bays used
in the original IBM PC. Most of todays drives
are half-height drives.
hand-held scanner Type of scanner that is
small enough to be held in your hand. Used to
digitize a relatively small image or artwork, it
consists of the controller, CCD, and light
source contained in a small enclosure with
wheels on it.
hard disk controller An expansion board
that contains the necessary circuitry to control
and coordinate a hard disk drive. Many hard
disk controllers are capable of managing more
than one hard disk, as well as floppy disks and
even tape drives.

graphics mode A mode of a video card that

allows the video card to display graphics.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

hard disk drive A storage device that uses a

set of rotating, magnetically coated disks called
platters to store data or programs. A typical
hard disk platter rotates at up to 7200 rpm, and
the read/write heads float on a cushion of air
from 10 to 25 millionths of an inch thick so
that the heads never come into contact with the
recording surface. The whole unit is hermetically sealed to prevent airborne contaminants
from entering and interfering with these close
tolerances. Hard disks range in capacity from a
few tens of megabytes to several gigabytes of
storage space; the bigger the disk, the more
important a well thought out backup strategy
becomes.
hard disk interfaces A connector on a motherboard that makes it possible to connect a
hard disk to the motherboard.
hard disk system A disk storage system containing the following components: the hard
disk controller, hard disk, and host adapter.
hard memory error A reproducible memory
error that is related to hardware failure.
hard page fault A hard page fault is when
data has been removed from memory and
added to the page file. When this data is
accessed, it is referred to as a hard page fault.
See page fault.

613

Hardware Compatibility List (HCL) An

HCL is a list (that is maintained and regularly
updated by Microsoft for each of its Windows
OSs) of all hardware currently known to be
compatible with a particular operating
system. Windows 98, NT, and 2000 all have
their own HCL.
Hardware Device Module (HDM) On a
NetWare network, this is a device driver that
allows the network operating system to interface with the peripheral.
hardware interrupt An interrupt or request
for service generated by a hardware device such
as a keystroke from the keyboard or a tick from
the clock. Because the processor may receive
several such signals simultaneously, hardware
interrupts are usually assigned a priority level
and processed according to that priority.
hardware ports

See I/O address.

head The electromagnetic device used to read

from and write to magnetic media such as hard
and floppy disks, tape drives, and compact
discs. The head converts the information read
into electrical pulses sent to the computer for
processing.
header Information that is attached to the
beginning of a network data frame.

hard reset A system reset made by pressing

the computers reset button or by turning the
power off and then on again.

heat sink A device that is attached to an electronic component that removes heat from the
component by induction. It is often a plate of aluminum or metal with several vertical fingers.

hardware All the physical electronic components of a computer system, including peripherals, printed-circuit boards, displays, and
printers.

hertz Abbreviated Hz. A unit of frequency

measurement; 1 hertz equals one cycle per
second.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

614

Glossary

hexadecimal Abbreviated hex. The base-16

numbering system that uses the digits 0 to 9, followed by the letters A to F (equivalent to the decimal numbers 10 through 15). Hex is a very
convenient way to represent the binary numbers
computers use internally, because it fits neatly
into the 8-bit byte. All of the 16 hex digits 0 to F
can be represented in 4 bits, and so two hex digits
(one digit for each set of 4 bits) can be stored in a
single byte. This means that 1 byte can contain
any one of 256 different hex numbers, from 0
through FF. Hex numbers are often labeled with
a lowercase h (for example, 1234h) to distinguish
them from decimal numbers.

8088 addresses. Programs that conform to the

extended memory specification can use this
memory as an extension of conventional
memory although only one program can use or
control HMA at a time.

Hierarchical PCI Bus A PCI Bus that has

been organized in a hierarchy. This is a method
of adding additional PCI Devices to a computer
system bus.

high-voltage power supply (HVPS) Provides the high voltages that are used during the
EP print process. This component converts
house AC currents into higher voltages that the
two corona assemblies can use.

high availability A term used to describe the

percentage of time that a server has been in service. If it has been available for a vast majority
of the time the system has been in service, it is
said to be a high availability device.

high-voltage probe A device used to drain

away voltage from a monitor before testing. It
is a pencil shaped device with a metal point and
a wire lead with a clip.

high-density disk A floppy disk with more

recording density and storage capacity than a
double-density disk.
high-level format The process of preparing
a floppy disk or a hard disk partition for use by
the operating system. In the case of DOS, a
high-level format creates the boot sector, the
file allocation table (FAT), and the root directory.
high memory area (HMA) In an IBMcompatible computer, the first 64K of extended
memory above the 1MB limit of 8086 and

High Voltage Differential (HVD) A method

of older SCSI communication with devices on
the SCSI chain. HVD supports throughput of 40
Mbytes/second at cable lengths of 25 meters.
These cards and drives are used in less than 5
percent of all implementations. They used to be
used in implementations where long cable runs
were necessary, especially in noisy areas.

HIMEM.SYS The DOS and Microsoft

Windows device driver that manages the use of
extended memory and the high memory area on
IBM-compatible computers. HIMEM.SYS not
only allows your application programs to access
extended memory, it oversees that area to prevent
other programs from trying to use the same space
at the same time. HIMEM.SYS must be loaded by
a DEVICE command in your CONFIG.SYS file;
you cannot use DEVICEHIGH.
HMA

See high memory area (HMA).

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

home page On the Internet, an initial

starting page. A home page may be related to a
single person, a specific subject, or a corporation and is a convenient jumping-off point to
other pages or resources.
horizontal cable Horizontal cable refers to
cable that services the same physical floor. (As
opposed to vertical cable, which is a backbone
that goes between floors or buildings.)
host The central or controlling computer in a
networked or distributed processing environment, providing services that other computers
or terminals can access via the network. Computers connected to the Internet are also
described as hosts, and can be accessed using
FTP, Telnet, Gopher, or a browser.
host adapter Translates signals from the
hard drive and controller to signals the computers bus can understand.
host name The name by which a computer is
known on a TCP/IP network. This name must
be unique within the domain that the machine
is in. In Windows 2000 the computer name is
always the same as the machines host name,
while in Windows 9x the two can be separate.
Hot Plug PCI The ability to plug a device into
a PCI slot without taking the PCI Bus off line.
hot spare This term is used in the discussion
of RAID devices. It is a spare drive in the RAID
array that is there only as a backup, in case
another drive fails. If there is the failure of a
drive, this drive is used to automatically rebuild
the data. Hot spares are kept online and do not
need user intervention to be activated.

615

are those that can be removed and replaced

without taking the computer or the system bus
out of service.
Host-to-PCI Bridge A system bus bridge that
manages the data being sent and received from
your PCI cards.
hub A connectivity device used to link several
computers together into a physical star
topology. They repeat any signal that comes in
on one port and copies it to the other ports.
HVPS See high-voltage power supply
(HVPS).
hybrid topology A mix of more than one
topology type used on a network.
Hypertext Transfer Protocol (HTTP) HTTP
is the protocol of the World Wide Web, and is
used to send and receive Web pages and other
content from an HTTP server (Web server).
HTTP makes use of linked pages, accessed via
hyperlinks, which are words or pictures that,
when clicked on, take you to another page.
I2O Messaging Layer Intelligent Input/
Output specification for the device driver architecture that is independent of both the device
being controlled and the host operating system.
I/O address Lines on a bus used to allow the
CPU to send instructions to the devices
installed in the bus slots. Each device is given its
own communication line to the CPU. These
lines function like one-way (unidirectional)
mailboxes.
I/O ports

See I/O address.

Hot Swap Changing devices without taking

the computer offline. Hot swappable devices

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

616

Glossary

IBM-compatible computer Originally, any

personal computer compatible with the IBM
line of personal computers. With the launch of
IBMs proprietary micro channel architecture
in the PS/2 line of computers, which replaced
the AT bus, two incompatible standards
emerged, and so the term became misleading.
Now, it is becoming more common to use the
term industry-standard computer when
referring to a computer that uses the AT or ISA
bus, and the term DOS computer to describe
any PC that runs DOS and is based on one of
the Intel family of chips.
IBM PC A series of personal computers based
on the Intel 8088 processor, introduced by IBM
in mid-1981. The PC was released containing
16K of memory, expandable to 64K on the
motherboard, and a monochrome video
adapter incapable of displaying bit-mapped
graphics. The floppy disk drive held 160K of
data and programs. There was no hard disk on
the original IBM PC; that came later with the
release of the IBM PC/XT.
IBM PS/2 A series of personal computers
using several different Intel processors, introduced by IBM in 1987. The main difference
between the PS/2 line and earlier IBM personal
computers was a major change to the internal
bus. Previous computers used the AT bus, also
known as industry-standard architecture, but
IBM used the proprietary micro channel architecture in the PS/2 line instead. Micro channel
architecture expansion boards will not work in
a computer using ISA. See IBM-compatible
computer.
IC

Icons On-screen graphics that act as doors

through which programs are started and therefore used to spawn windows. They are shortcuts that allow a user to open a program or a
utility without knowing where that program is
or how it needs to be configured.
IDE Acronym for integrated drive electronics.
A hard disk technology that can connect multiple drives together. These drives integrate the
controller and drive into one assembly. This
makes them very inexpensive. Because of this,
IDE drives are the most commonly used disk
technology installed in computers today.
Identify Drive A controller-level command
designed to identify the device on the disk subsystem chain and make sure it is operating
properly.
IEEE-1394 A high-speed digital interface
most commonly used to transfer data between
computers and digital video cameras. It has a
maximum data transfer rate of over 400MBps.
illegal operation error A Windows error
that occurs when a program does something
that Windows wasnt expecting or doesnt
know how to do.
impact printer Any printer that forms an
image on paper by forcing a character image
against an inked ribbon. Dot-matrix, daisywheel, and line printers are all impact printers,
whereas laser printers are not.
incremental backup A backup of a hard
disk that consists of only those files created or
modified since the last backup was performed.

See integrated circuit (IC).

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

incremental zone transfer In Domain

Name Service (DNS), an incremental zone
transfer is where only the parts of the database
that have been changed are copied from one
DNS database server to another.
industry-standard architecture

See ISA.

INI file Text file that is created by an installation program when a new Windows application is installed. INI files contain settings for
individual Windows applications as well as for
Windows itself.
initialization commands A set of commands
sent to a modem to prepare it to function.
inoculating The process of protecting a
computer system against virus attacks by
installing antivirus software.
input/output addresses

See I/O address.

Institute of Electrical and Electronic Engineers (IEEE) A group of engineering and

electronics professionals that is based in the
United States but has membership all over the
world. The IEEE focuses on standards for electrical, electronic, computer engineering, and
other science-related matters.
integrated circuit (IC) Also known as a chip.
A small semiconductor circuit that contains
many electronic components.
integrated drive electronics

See IDE.

Integrated Services Digital Network See

ISDN.
integrated system boards A system board
that has most of the computers circuitry
attached, as opposed to having been installed as
expansion cards.

617

Intel OverDrive OverDrive chips boost

system performance by using the same clock
multiplying technology found in the Intel
80486DX-2 and DX4 chips. Once installed, an
OverDrive processor can increase application
performance by an estimated 40 to 70 percent.
intelligent hub A class of hub that can be
remotely managed on the network.
Intelligent Input/Output (I2O) Intelligent
Input/Output specification for the device driver
architecture that is independent of both the
device being controlled and the host operating
system.
interface Any port or opening that is specifically designed to facilitate communication
between two entities.
interface software The software for a particular interface that translates software commands into commands that the printer can
understand.
interlacing A display technique that uses two
passes over the monitor screen, painting every
other line on the screen the first time and then
filling in the rest of the lines on the second pass.
It relies on the physiological phenomenon
known as persistence of vision to produce the
effect of a continuous image.
interleaved memory A method of organizing the RAM memory address in such a way
as to reduce the number of wait states necessary
to access the information. With interleaved
memory, adjacent locations are stored in different rows of chips so that the processor can
access bytes more efficiently.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

618

Glossary

interleaving Interleaving involves skipping

sectors to write the data, instead of writing
sequentially to every sector. This evens out the
data flow and allows the drive to keep pace
with the rest of the system. Interleaving is given
in ratios. If the interleave is 2:1, the disk skips 2
minus 1, or 1 sector, between each sector it
writes (it writes to one sector, skips one sector,
then writes to the next sector following). Most
drives today use a 1:1 interleave, because
todays drives are very efficient at transferring
information.
International Organization for Standardization (ISO) An international standard-making
body, based in Geneva, that establishes global
standards for communications and information exchange.
Internet The Internet (Net) is the global TCP/
IP network that now extends into nearly every
office and school. The World Wide Web is the
most visible part of the Internet, but e-mail,
newsgroups, and FTP (to name just a few) are
also important parts of the Internet.
Internet address An IP or domain address
which identifies a specific node on the Internet.
Internet Control Message Protocol
(ICMP) A member of the TCP/IP protocol
suite that operates at Layer 3 (Network layer).
ICMP provides error correction and other
information that is needed in the processing of
an IP packet.
Internet Group Membership Protocol
(IGMP) A protocol of the TCP/IP protocol
suite that is used by IP hosts to report their host
group membership to any immediately neighboring multicast routers.

Internet Information Server (IIS) IIS is the

Microsoft Web Server that runs on Windows
NT Servers and also on Windows 2000 Servers.
Internetwork Packet Exchange (IPX)
Transport protocol used by Novell NetWare
2.x, 3.x, and 4.x implementations for routing
packets between LANs.
Internet Protocol

See IP.

Internet Service Provider (ISP) An ISP is a

company that provides Internet access for
users. Generally ISPs are local or regional companies that provide Internet access and e-mail
addresses to users.
internetwork Any TCP/IP network that
spans router interfaces is considered to be an
internetwork. This means that anything from a
small office with two subnets to the Internet
itself can be described as an internetwork.
interrupt A signal to the processor generated
by a device under its control (such as the system
clock) that interrupts normal processing. An
interrupt indicates that an event requiring the
processors attention has occurred, causing the
processor to suspend and save its current
activity and then branch to an interrupt service
routine. This service routine processes the
interrupt (whether it was generated by the
system clock, a keystroke, or a mouse click)
and when its complete, returns control to the
suspended process. In the PC, interrupts are
often divided into three classes: internal hardware, external hardware, and software interrupts. The Intel 80x86 family of processors
supports 256 prioritized interrupts, of which
the first 64 are reserved for use by the system
hardware or by DOS.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

interrupt request (IRQ) A hardware interrupt signals that an event has taken place that
requires the processors attention, and may
come from the keyboard, the input/output
ports, or the systems disk drives. In the PC, the
main processor does not accept interrupts from
hardware devices directly; instead interrupts
are routed to an Intel 8259A Programmable
Interrupt Controller. This chip responds to
each hardware interrupt, assigns a priority, and
forwards it to the main processor.
interrupt request (IRQ) lines Hardware
lines that carry a signal from a device to the
processor.
IP Abbreviation for Internet Protocol. The
underlying communications protocol on which
the Internet is based. IP allows a data packet to
travel across many networks before reaching its
final destination.
IP address In order to communicate on a
TCP/IP network, each machine must have a
unique IP address. This address is in the form
x.x.x.x where x is a number from 0 to 255.
IPCONFIG Used on Windows 2000 to view
current IP configuration information and to
manually request updated information from a
DHCP server.
IPSec Internet Protocol Security protocols. A
developing security standard that works at the
Network layer or at the Packet Processing layer
of network communications. IPSec is usually
discussed when the subject is Virtual Private
Networks (VPNs).
IRQ

See interrupt request (IRQ).

619

ISA Abbreviation for industry-standard

architecture. The 16-bit bus design was first
used in IBMs PC/AT computer in 1984. ISA
has a bus speed of 8MHz and a maximum
throughput of 8 megabytes per second. EISA is
a 32-bit extension to this standard bus.
ISDN Abbreviation for Integrated Services
Digital Network. A worldwide digital communications network emerging from existing telephone services, intended to replace all current
systems with a completely digital transmission
system. Computers and other devices connect
to ISDN via simple, standardized interfaces,
and when complete, ISDN systems will be
capable of transmitting voice, video, music,
and data.
Journaled File System
on OS/2 servers.
joystick port

The file system used

See game port.

jumper A small plastic and metal connector that

completes a circuit, usually to select one option
from a set of several user-definable options.
Jumpers are often used to select one particular
hardware configuration rather than another.
kernel file Windows core component that is
responsible for managing Windows resources
and running applications.
kernel thread The core of the operating
system. This is the part of the operating system
that operates the memory, files, and peripherals, as well as maintaining the time and date,
launching applications, and allocating system
resources.
keyboard, video, mouse (KVM) Switches
designed to share a common keyboard, monitor,
and mouse among several computers.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

620

Glossary

kilobit Abbreviated Kb or Kbit. 1024 bits

(binary digits).

just one of many factors that influence disk

access speeds.

kilobits per second Abbreviated Kbps. The

number of bits, or binary digits, transmitted
every second, measured in multiples of 1024
bits per second. Used as an indicator of communications transmission rate.

Layer 2 Tunneling Protocol (L2TP) Also

referred to as Layer 2 Tunneling Protocol.
L2TP is an extension of Point to Point Tunneling Protocol (PPTP), which is used by
Internet Service Providers to establish Virtual
Private Networks (VPNs). L2TP is actually a
merger of two other protocols, PPTP from
Microsoft and L2F from Cisco Systems.

kilobyte
bytes.

Abbreviated K, KB, or Kbyte. 1024

L1 Cache (L1) Any cache memory that is

integrated into the CPU.

LCD

L2 Cache (L2) Any cache memory that is

external to the CPU.

LCD monitor A monitor that uses liquid

crystal display technology. Many laptop and
notebook computers use LCD displays because
of their low power requirements.

LAN

See local area network (LAN).

See liquid crystal display (LCD).

LAN switches A form of concentrator or

hub, but with intelligence. A LAN switch will
create a virtual circuit between the two communicating hosts for the duration of the conversation. Used in Ethernet networks to eliminate
the congestion normally found in technology
based on CSMA/CD.

least significant bit (LSB) In a binary

number, the lowest-order bit. That is, the rightmost bit. So, in the binary number 0001, the 1
is the least significant bit.

Iaser printer A generic name for a printer

that uses the electrophotographic (EP) print
process.

legacy A component that is still functional

but is out of date.

laser scanner The assembly in an EP process

printer that contains the laser. This component
is responsible for writing the image to the EP
drum.
latency The time that elapses between issuing
a request for data and actually starting the data
transfer. In a hard disk, this translates into the
time it takes to position the disks read/write
head and rotate the disk so that the required
sector or cluster is under the head. Latency is

LED page printer A type of EP process

printer that uses a row of LEDs instead of a
laser to expose the EP drum.

letter quality (LQ) A category of dot-matrix

printer that can print characters that look very
close to the quality a laser printer might produce.
level of risk The process in disaster recovery
where assets are reviewed and decisions are
made as to how much risk the resource is actually being subjected to. Once the level of risk
has been assessed, the disaster recovery team
can determine appropriate levels of protection.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

621

Lightweight Directory Access Protocol

(LDAP) A member of the TCP/IP protocol
suite that extracts information from a hierarchical directory, such as an X.500 standard
database. This gives users a tool to use to locate
resources on the network.

(Extended Industry Standard Architecture) definitions. Local bus can achieve a maximum
data rate of 133 megabytes per second with a
33MHz bus speed, 148 megabytes per second
with a 40MHz bus, or 267 megabytes per
second with a 50MHz bus.

Link Control Facility (LCF) In Fibre Channel

communications the Link Control Facility
(LCF) handles the logical and physical control
of the link at the firmware level.

local resources Files or folders that are

physically located on the machine the user is
sitting at are referred to as local to that user.
Windows 2000 has the ability to enforce local
security, while Windows 9x does not.

liquid crystal display (LCD) A display technology common in portable computers that
uses electric current to align crystals in a special
liquid. The rod-shaped crystals are contained
between two parallel transparent electrodes,
and when current is applied, they change their
orientation, creating a darker area. Many LCD
screens are also backlit or side-lit to increase
visibility and reduce the possibility of eyestrain.

Logical Block Addressing (LBA) In IDE

drives, Logical Block Addressing (LBA) is a
way for the system to access IDE hard disk
drives that exceed 504MB in size. Before LBA,
limitations in the ROM BIOS prevented the
DOS FDISK command from using an entire
hard drive if it was greater than 504MB.

Load Balancing The process where communications are spread between various networking components to manage the load and
spread out the work. This can be done at the
network card level, at the hard drive level, or
even at the application server level.

logic board The sturdy sheet or board to

which all other components on the computer
are attached. These components consist of the
CPU, underlying circuitry, expansion slots,
video components, and RAM slots, just to
name a few. Also known as a motherboard or
planar board.

local area network (LAN) A group of computers and associated peripherals connected by
a communications channel capable of sharing
files and other resources between several users.

logical drive Created within an extended

partition, a logical drive is used to organize
space within the partition, which can be
accessed through the use of a drive letter.

local bus A PC bus specification that allows

peripherals to exchange data at a rate faster
than the 8 megabytes per second allowed by the
ISA (Industry Standard Architecture) and the
32 megabytes per second allowed by the EISA

logical memory The way memory is organized so it can be accessed by an operating

system.
logical topology Topology that defines how
the data flows in a network.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

622

Glossary

logon The process of logging on submits

your username and password to the network
and gives you the network credentials you will
use for the rest of that session. Users can either
log on to a workgroup or to a network security
entity (such as the Active Directory).
low-level format The process that creates
the tracks and sectors on a blank hard disk or
floppy disk; sometimes called the physical
format. Most hard disks are already low-level
formatted; however, floppy disks receive both a
low- and a high-level format (or logical format)
when you use the DOS or OS/2 command
FORMAT.
Low Voltage Differential (LVD) In SCSI,
LVD is a standard from the X3T10 SCSI Committee that is used to double the data
throughput of SCSI devices. LVD uses 2.2 volt
logic. This gives the SCSI interface the advantage of noise rejection and longer cable lengths.
LPTx ports In DOS, the device name used to
denote a parallel communications port, often
used with a printer. DOS supports three parallel
ports: LPT1, LPT2, and LPT3, and OS/2 adds
support for network ports LPT4 through LPT9.
Low Voltage Differential Multimode
Single-Ended (LVD/MSE) This is an implementation of SCSI that will automatically
switch between the LVD and the single-ended
mode. When a single-ended device is connected
to a multimode LVD/MSE bus, the entire bus
switches to the single-ended mode. Otherwise
LVD/MSE devices operate in the LVD mode.
magneto-optical (MO) drives An erasable,
high-capacity, removable storage device similar
to a CD-ROM drive. Magneto-optical drives
use both magnetic and laser technology to write

data to the disk and use the laser to read that

data back again. Writing data takes two passes
over the disk, an erase pass followed by the
write pass, but reading can be done in just one
pass and, as a result, is much faster.
mail server A server that is dedicated to handling the functions of electronic mail.
main memory

See RAM.

main motor A printer stepper motor that is

used to advance the paper.
maintenance log A written record of any
maintenance that has been done on a particular
computer. This can include the addition or
removal of any hardware and software components and the configuration of those components.
Management Information Base (MIB) This
is a database that contains the description of
network objects that can be managed using
Simple Network Management Protocol
(SNMP). The format of the MIB is defined as
part of the SNMP specification, and all other
variations are extensions of this basic MIB.
Management Information Format (MIF)
Database This is a database that contains the
description of desktop or workstation objects
that can be managed using Desktop Management
Interface (DMI). The format of the MIF is defined
as part of the DMI specification, and all other
variations are extensions of this basic MIF.
Management Interface (MI) The interface
for managing a desktop using Desktop Management Interface (DMI).

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

Manchester encoding An electrical specification that is a synchronous clock encoding

technique used by the OSI physical layer to
encode the clock and data of a synchronous bit
stream.
master

See master drive.

master drive The primary drive in an IDE

master/slave configuration.
math coprocessor A processor that speeds up
the floating decimal point calculations that are
needed in algebra and statistical calculations.
MCA MCA is incompatible with expansion
boards that follow the earlier 16-bit AT bus
standard, physically because the boards are
about 50 percent smaller and electronically as
the bus depends on more proprietary integrated
circuits. MCA was designed for multiprocessing, and it also allows expansion boards to
identify themselves, thus eliminating many of
the conflicts that arose through the use of
manual settings in the original bus.
Mean Time Between Failure (MTBF) The
average amount of time that a device will function without failing. Usually measured in
10,000 hour or 100,000 hour increments.
Media Independent Interface (MII)
connector A connector used to connect fiber
optic cables.
megabit (Mbit) Usually 1,048,576 binary
digits or bits of data. Often used as equivalent
to 1 million bits.
megabits per second (Mbps) A measurement of the amount of information moving

623

across a network or communications link in 1

second, measured in multiples of 1,048,576 bits.
megabyte (MB) Usually 1,048,576 bytes.
Megabytes are a common way of representing
computer memory or hard-disk capacity.
megahertz (MHz) One million cycles per
second. A processors clock speed is often
expressed in MHz. The original IBM PC operated an 8088 running at 4.77MHz, while the
more modern Pentium processor runs at speeds
of up to 1000MHz and higher.
Mega Transfer (MT) A unit of measure
referring to the rate of signals on the SCSI interface regardless of the width of the SCSI bus.
memory The primary random access
memory (RAM) installed in the computer. The
operating system copies application programs
from disk into memory, where all program execution and data processing takes place; results
are written back out to disk again. The amount
of memory installed in the computer can determine the size and number of programs that it
can run, as well as the size of the largest data
file.
memory address The exact location in
memory that stores a particular data item or
program instruction.
Memory Expansion Technology
(MXT) MXT is a hardware implementation
that automatically stores frequently accessed
data and instructions close to a computer's
microprocessors so they can be accessed
immediately, thereby significantly improving
performance. It was designed by IBM.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

624

Glossary

memory map The organization and allocation of memory in a computer. A memory map
will give an indication of the amount of
memory used by the operating system and the
amount remaining for use by applications.

Microsoft Diagnostics See MSD

(Microsoft Diagnostics).

memory optimization The process of

making the most possible conventional
memory available to run DOS programs.

Microsoft Management Console (MMC) A

series of management GUIs for Windows 2000
allowing the administrator to manage objects
stored in the Active Directory.

memory refresh An electrical signal that

keeps the data stored in memory from
degrading.
mesh topology Type of logical topology
where each device on a network is connected to
every other device on the network. This
topology uses routers to search multiple paths
and determine the best path.
Messaging Application Programming Interface (MAPI) The MAPI interface is used to
control how Windows interacts with messaging
applications such as e-mail programs. MAPI
makes most of the functions of e-mail transparent and allows programmers to just write
the application, not the whole messaging
system.
MFM encoding See modified frequency
modulation (MFM) encoding.
Micro Channel Architecture (MCA)
MCA.

See

microcode Microcode is a programming

technique that is used to install or modify a
device by altering programmable read-only
memory (PROM) or erasable programmable
read-only memory (EPROM).

Microsoft Disk Operating System

MS-DOS.

See

Microsoft Systems Management Server

(SMS) A suite of management utilities and
SNMP compliant devices that assist in the management of a Microsoft Windows NT network.
mirroring

See disk mirroring.

modem Contraction of modulator/demodulator, a device that allows a computer to

transmit information over a telephone line. The
modem translates between the digital signals
that the computer uses and analog signals suitable for transmission over telephone lines.
When transmitting, the modem modulates the
digital data onto a carrier signal on the telephone line. When receiving, the modem performs the reverse process and demodulates the
data from the carrier signal.
modified frequency modulation (MFM)
encoding The most widely used method of
storing data on a hard disk. Based on an earlier
technique known as frequency modulation
(FM) encoding, MFM achieves a two-fold
increase in data storage density over standard
FM recording, but it is not as efficient a space
saver as run-length limited encoding.
Molex connector
power connector.

Copyright 2001 SYBEX, Inc., Alameda, CA

See standard peripheral

www.sybex.com

Glossary

monitor A video output device capable of

displaying text and graphics, often in color.
monochrome monitor A monitor that can
display text and graphics in one color only. For
example, white text on a green background or
black text on a white background.
most significant bit (MSB) In a binary
number, the highest-order bit. That is, the leftmost bit. In the binary number 10000000, the 1
is the most significant bit.
motherboard The main printed circuit
board in a computer that contains the central
processing unit, appropriate coprocessor and
support chips, device controllers, memory, and
also expansion slots to give access to the computers internal bus. Also known as a logic
board or system board.
mouse A small input device with one or more
buttons used as for pointing or drawing. As you
move the mouse in any direction, an on-screen
mouse cursor follows the mouse movements;
all movements are relative. Once the mouse
pointer is in the correct position on the screen,
you can press one of the mouse buttons to initiate an action or operation; different user
interfaces and file programs interpret mouse
clicks in different ways.
MSBACKUP A DOS program that allows the
user to make backup copies of all the programs
and data stored on the hard disk. This program
is menu-driven and allows the user to set up
options that can be used each time you back up
the hard drive.

625

MSD (Microsoft Diagnostics) Program that

allows the user to examine many different
aspects of a systems hardware and software
setup.
MS-DOS Acronym for Microsoft Disk Operating System. MS-DOS, like other operating
systems, allocates system resources (such as
hard and floppy disks, the monitor, and the
printer) to the applications programs that need
them. MS-DOS is a single-user, single-tasking
operating system, with either a command-line
interface or a shell interface.
multimedia A computer technology that displays information by using a combination of
full-motion video, animation, sound, graphics,
and text with a high degree of user interaction.
multimeter Electronic device used to measure and test ohms, amperes, and volts.
multiplexer A network device that combines
multiple data streams into a single stream for
transmission. Multiplexers can also break out
the original data streams from a single, multiplexed stream.
Multi-Purpose Internet Mail Extensions
(MIME) An e-mail specification that allows
for the transfer of file formats other than text.
This is the protocol that allows you to attach
pictures and other formatted files to an e-mail
message and have the attachments come
through.
multipurpose server A server that has more
than one use. For example, a multi-purpose server
can be both a file server and a print server.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

626

Glossary

multistation access unit (MAU) The central device in a Token Ring network that provides both the physical and logical connections
to the stations.

NetBEUI Abbreviation for NetBIOS

Extended User Interface. A network device
driver for the transport layer supplied with
Microsofts LAN Manager.

multisync monitor A monitor designed to

detect and adjust to a variety of different input
signals. By contrast, a fixed-frequency monitor
must receive a signal at one specific frequency.

NetBIOS Acronym for Network Basic Input/

Output System. In networking, a layer of software, originally developed in 1984 by IBM and
Sytek, that links a network operating system
with specific network hardware. NetBIOS provides an application program interface (API)
with a consistent set of commands for
requesting lower-level network services to
transmit information from node to node.

multitasking A feature of an operating

system that allows more than one program to
run simultaneously.
multithreading The ability of a program to
send multiple tasks to the processor at the same
time. This allows an application to execute
more quickly, but it requires the support of a
multithreaded operating system.
N_port Node port. A port on a computer,
disk drive, etc., through which the device does
its Fibre Channel communication.
NL_port Node-Loop port. An N_Port that
can operate on the Arbitrated Loop topology.
near letter quality (NLQ) A category of dotmatrix printer that can come close to the
quality of a laser printer, but still is lacking
somewhat in print quality.
nearest active upstream neighbor
(NAUN) In a Token Ring network, each node
is associated with two other nodes, including
the NAUN. The NAUN passes frames and the
token to this node.

NetBIOS Extended User Interface

NetBEUI.

See

NetWare Directory Services (NDS) X.500

compliant directory that is used on NetWare
4.x and 5.x networks. Also referred to as the
e-Directory.
NetWare Loadable Modules (NLMs) Components of a NetWare 3.x, 4.x, and 5.x server
that allow for configuration of the server and
the addition of services.
Network Address Translation (NAT) NAT
is the translation of an Internet Protocol
address that is used within one network to a
different IP address known within another network. One network is designated the private
network and the other is the public network.
Typically, a company maps its local private network addresses to one or more global public IP
addresses and unmaps the public IP addresses
on incoming packets back into private IP
addresses.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

network A group of computers and associated peripherals connected by a communications channel capable of sharing files and other
resources between several users. A network can
range from a peer to peer network (which connects a small number of users in an office or
department) to a local area network (which
connects many users over permanently
installed cables and dial-up lines) or to a wide
area network (that connects users on several
different networks spread over a wide geographic area).
network adapter In order to access network
resources, a physical connection to the network
must be made. This is generally done through
the network adapter, which is expansion hardware designed to interface with the network.
Network Basic Input/Output System
NetBIOS.

See

network client software The software that

enables a computer to communicate on the
network.
network interface card (NIC) In networking, the PC expansion board that plugs
into a personal computer or server and works
with the network operating system to control
the flow of information over the network. The
network interface card is connected to the network cabling (twisted-pair, coaxial or fiberoptic cable), which in turn connects all the network interface cards in the network.
Network layer The third of seven layers of
the International Standards Organizations
Open Systems Interconnection (ISO/OSI)
model for computer-to-computer communications. The Network layer defines protocols for

627

data routing to ensure that the information

arrives at the correct destination node.
network operating system (NOS)
NOS.

See

network security provider In a network

environment, it is often easier to manage the
network by having centralized user ID and
password storage. Examples of this type of centralized system are Windows 2000s Active
Directory or NetWares NDS.
network segmentation The process of
dividing up a network into smaller parts.
This is done for a variety of reasons, including
network management and bandwidth
management.
NIC

See network interface card (NIC).

node In communications, any device

attached to the network.
nonconductor Any material that does not
conduct electricity.
nondedicated server A computer that can
be both a server and a workstation. In practice,
by performing the functions of both server and
workstation, this type of server does neither
function very well. Nondedicated servers are
typically used in peer-to-peer networks.
nonintegrated system boards A type of
motherboard where the various subsystems
(video, disk access, etc.) are not integrated into
the motherboard, but rather placed on expansion cards that can be removed and upgraded.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

628

Glossary

non-interlaced Describes a monitor in which

the display is updated (refreshed) in a single
pass, painting every line on the screen. Interlacing takes two passes to paint the screen,
painting every other line on the first pass, and
then sequentially filling in the other lines on the
second pass. Non-interlaced scanning, while
more expensive to implement, reduces
unwanted flicker and eyestrain.
non-routable protocol A protocol like netBEUI that cannot be passed from one network
segment to another through the use of routing
protocols.
NOS (Network Operating System) Software that runs on the server and controls and
manages the network. The NOS controls the
communication with resources and the flow of
data across the network.
notebook computer A small portable computer, about the size of a computer book, with
a flat screen and a keyboard that fold together.
A notebook computer is lighter and smaller
than a laptop computer. Some models use flash
memory rather than conventional hard disks
for program and data storage, while other
models offer a range of business applications in
ROM. Many offer PCMCIA expansion slots
for additional peripherals such as modems, fax
modems, or network connections.
Novell Distributed Print Services (NDPS)
NDPS is the way that NetWare implements
shared printing on a NetWare network.
Novell Storage Services (NSS) NSS is the
storage management system on a NetWare network.

NTFS The NT File System (NTFS) was created to provide enhanced security and performance for the Windows NT operating system,
and it has been adopted and improved upon by
Windows 2000. NTFS provides Windows
2000 with local file security, file auditing, compression, and encryption options. It is not compatible with Windows 9x or DOS.
null modem A short RS-232-C cable that
connects two personal computers so that they
can communicate without the use of modems.
The cable connects the two computers serial
ports, and certain lines in the cable are crossed
over so that the wires used for sending data by
one computer are used for receiving data by the
other computer and vice versa.
numeric keypad A set of keys to the right of
the main part of the keyboard, used for
numeric data entry.
odd parity A technique that counts the
number of 1s in a binary number and, if the
number of 1s total is not an odd number, adds
a digit to make it odd. See also parity.
ohm

Unit of electrical resistance.

Open Shortest Path First (OSPF) OSPF is a

protocol that is used to route packets within
larger networks. It is used instead of other protocols like the Routing Information Protocol
(RIP). RIP is an older routing protocol that is
installed in many of today's corporate networks. Like RIP, OSPF is designated by the
Internet Engineering Task Force (IETF) as one
of several Interior Gateway Protocols.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

Open Systems Interconnection (OSI)

model See OSI (Open Systems Interconnection) model.
operating system (OS) The software
responsible for allocating system resources,
including memory, processor time, disk space,
and peripheral devices such as printers,
modems, and the monitor. All application programs use the operating system to gain access
to these system resources as they are needed.
The operating system is the first program
loaded into the computer as it boots, and it
remains in memory at all times thereafter.
optical disk A disk that can be read from and
written to, like a fixed disk but, like a CD, is
read with a laser.
optical drive A type of storage drive that
uses a laser to read from and write to the
storage medium.
optical mouse A mouse that uses a special
mouse pad and a beam of laser light. The beam
of light shines onto the mouse pad and reflects
back to a sensor in the mouse. Special small
lines crossing the mouse pad reflect the light
into the sensor in different ways to signal the
position of the mouse.
optical scanner

See scanner.

optical touch screen A type of touch screen

that uses light beams on the top and left side
and optical sensors on the bottom and right
side to detect the position of your finger when
you touch the screen.
option diskette A diskette that contains the
device-specific configuration files for the device
being installed into a MCA bus computer.

629

opto-mechanical mouse Type of mouse

that contains a round ball that makes contact
with two rollers. Each roller is connected to a
wheel that has small holes in it. The wheel
rotates between the arms of a U-shaped mechanism that holds a light on one arm and an
optical sensor on the other. As the wheels
rotate, the light flashes coming through the
holes indicate the speed and direction of the
mouse, and these values are transmitted to the
computer and the mouse control software.
OSI (Open Systems Interconnection)
model A protocol model, developed by the
International Standards Organization (ISO),
that was intended to provide a common way of
describing network protocols. This model
describes a seven-layered relationship between
the stages of communication. Not every protocol maps perfectly to the OSI model, as there
is some overlap within some the layers of some
protocols.
OS Services Module (OSM) In an I20
system OSM is what provides the interface
between the host operating system and the I2O
Message Layer. In the split driver module, the
OSM represents the portion of the driver that
interfaces to host specific Application Programming Interfaces (APIs), translating them to a
neutral message based format that is then sent
to an Hardware Device Module (HDM) for
processing.
The OSM translates requests from the host
operating system into messages that can be dispatched to the appropriate HDM for processing. HDM information is forwarded back
to host operating system through the OSM via
the I2O Message Layer.
Packet Internet Grouper (PING)

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

See PING.

630

Glossary

Packet Receive Buffer In NetWare, a packet

receive buffer is an area of memory that is
reserved to hold incoming requests from the
network. The requests are in the form of
packets.
paddleboard
troller card.

Another name for an IDE con-

page description language (PDL)

Describes the whole page being printed. The
controller in the printer interprets these commands and turns them into laser pulses or firing
print wires.
page fault When your computer is looking
for information in RAM and cannot find it, it
then looks to the page file. This is referred to as
a page fault.
page frame The special area reserved in
upper memory that is used to swap pages of
memory into and out of expanded memory.
page printers Type of printer that handles
print jobs one page at a time instead of one line
at a time.
paged mode memory A specially designed
dynamic RAM that supports access to sequential memory locations with a reduced cycle
time. Page mode RAM improves the execution
speed of code because the code tends to execute
sequentially through memory.
pages 16K chunks of memory used in
expanded memory.
paging The process of swapping memory to
an alternate location, such as to and from a
page frame in expanded memory or to and
from a swap file.

Palo Alto Research Center (PARC) A

facility in Palo Alto, California, owned by
Xerox, where many of the discoveries that led
to early networking were made.
paper pickup roller A D-shaped roller that
rotates against the paper and pushes one sheet
into the printer.
paper registration roller A roller in an EP
process printer that keeps paper movement in
sync with the EP image formation process.
paper transport assembly The set of
devices that moves the paper through the
printer. It consists of a motor and several rubberized rollers that each perform a different
function.
Parallel ATA The type of interface used by
some IDE devices. This method of data transfer
has been referred to as a potential bottleneck.
Several groups are attempting to replace it with
Serial ATA.
parallel port An input/output port that manages information 8 bits at a time, often used to
connect a parallel printer.
parallel processing A processor architecture
where a processor essentially contains two processors in one. The processor can then execute
more than one instruction per clock cycle.
parity Parity is a simple form of error checking
used in computers and telecommunications.
Parity works by adding an additional bit to a
binary number and using it to indicate any
changes in that number during transmission.
partition A portion of a hard disk that the
operating system treats as if it were a separate
drive.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

partition table In DOS, an area of the hard

disk containing information on how the disk is
organized. The partition table also contains
information that tells the computer which
operating system to load; most disks will contain DOS, but some users may divide their hard
disk into different partitions, or areas, each
containing a different operating system. The
partition table indicates which of these partitions is the active partition, the partition that
should be used to start the computer.
passive hub Type of hub that electrically
connects all network ports together. This type
of hub is not powered.
passive-matrix screen An LCD display
mechanism that uses a transistor to control
every row of pixels on the screen. This is in
sharp contrast to active-matrix screens, where
each individual pixel is controlled by its own
transistor.
Passive Termination SCSI termination technology. This method is a very simple termination method that works best with four or fewer
devices on a SCSI daisy chain.
password In order to identify themselves on
the network, each user must provide two credentialsa username and a password. The
username says, This is who I am, and the
password says, And heres proof! Passwords
are case sensitive and should be kept secret
from other users on the network.
path When referring to a file on a computers
hard drive, the path is used to describe where it
exists within the directory structure. If a file is
on the D drive in a folder named TEST, its path
is d:\test\.

631

PC Card A PC Card, also known as a

PCMCIA card or a credit card adapter is a
peripheral device that uses the PCMCIA specification. These have the advantage of being
small, easy to use and fully plug-and-play compliant.
PC Card slot An opening in the case of a portable computer intended to receive a PC Card;
also known as a PCMCIA slot.
PC Card Socket Services
PCB

See socket services.

See printed-circuit board (PCB).

PC-DOS 1.0 Microsofts Disk Operating

System is generally referred to as MS-DOS.
When it was packaged with IBMs personal
computers, though, DOS was modified slightly
and was called PC-DOS.
PCI Abbreviation for Peripheral Component
Interconnect. A specification introduced by
Intel that defines a local bus that allows up to
10 PCI-compliant expansion cards to be
plugged into the computer. One of these 10
cards must be the PCI controller card, but the
others can include a video card, network interface card, SCSI interface, or any other basic
input/output function. The PCI controller
exchanges information with the computers
processor as 32- or 64-bits and allows intelligent PCI adapters to perform certain tasks concurrently with the main processor by using bus
mastering techniques.
PCI Bridge A device to connect several PCI
Buses together to extend the functionality of
the main system bus.
PCI Hot Plug A technology that allows
devices to be plugged directly into a PCI Bus
without taking the computer out of service.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

632

Glossary

PCI-to-PCI Bridge A set of chips that are

used on motherboards and add-in cards to
allow a system designer to increase data
throughput in a variety of data intensive server,
workstation, and high-end PC applications,
including networking and data storage.
PCMCIA Abbreviation for PC Memory Card
International Association. Expansion cards
developed for this standard are now called PC
Cards.

peripheral Any hardware device attached to

and controlled by a computer, such as a monitor, keyboard, hard disk, floppy disk, CDROM drives, printer, mouse, tape drive, and
joystick.
Peripheral Component Interconnect
PCI.

See

peer to peer network Describes a network

where the computers act as both workstations
and servers and where there is no centralized
administration or control.

permanent swap file A permanent swap file

allows Microsoft Windows to write information to a known place on the hard disk, which
enhances performance over using conventional
methods with a temporary swap file. The Windows permanent swap file consists of a large
number of consecutive contiguous clusters; it is
often the largest single file on the hard disk, and
of course this disk space cannot be used by any
other application.

Personal System/2 (PS/2) A type of personal computer and server designed and marketed by IBM.

PGA (Pin Grid Array) A type of IC package

that consists of a grid of pins connected to a
square, flat package.

Pentium The Pentium represents the evolution of the 80486 family of microprocessors
and adds several notable features, including 8K
instruction code and data caches, built-in
floating-point processor and memory management unit, as well as a superscalar design and
dual pipelining that allow the Pentium to execute more than one instruction per clock cycle.

photosensitive drum

Peer PCI Bus Adding a second PCI Bus to a

system and Load Balancing between them.

Pentium Pro The 32-bit Pentium Pro (also

known as the P6) has a 64-bit data path
between the processor and cache and is capable
of running at clock speeds up to 200MHz.
Unlike the Pentium, the Pentium Pro has its secondary cache built into the CPU itself, rather
than on the motherboard, meaning that it
accesses cache at internal speed, not bus speed.

physical drive
drive.

See EP drum.

Another term for a hard disk

Physical layer The first and lowest of the

seven layers in the International Standards
Organizations Open Systems Interconnection
(ISO/OSI) model for computer-to-computer
communications. The Physical layer defines the
physical, electrical, mechanical, and functional
procedures used to connect the equipment.
physical topology A description that identifies how the cables on a network are physically
arranged.
pickup roller

Copyright 2001 SYBEX, Inc., Alameda, CA

See paper pickup roller.

www.sybex.com

Glossary

Pin Grid Array

See PGA (Pin Grid Array).

PING PING is a utility used to send a short

message to another computer on a TCP/IP network. PING can be useful to test connectivity
between networks or to see if a particular
machine is communicating with the network.
pixel Contraction of picture element. The
smallest element that display software can use
to create text or graphics. A display resolution
described as being 640 480 has 640 pixels
across the screen and 480 down the screen, for
a total of 307,200 pixels. The higher the
number of pixels, the higher the screen resolution. A monochrome pixel can have two values,
black or white, and this can be represented by 1
bit as either 0 or 1. At the other end of the scale,
true color, capable of displaying approximately
16.7 million colors, requires 24 bits of information for each pixel.
planar board

See motherboard.

planned outage A planned period of time

where a system will be taken out of service for
routine maintenance or to add additional components.
platform An operating system (OS) is the
basic software that runs on a computer, and it
is the base on which all other software sits. As
such the OS is the platform that applications
and utilities run on.
plenum The space between a dropped ceiling
and the actual ceiling in a building.
Plug and Play (PnP) A standard that defines
automatic techniques designed to make PC
configuration simple and straightforward.

633

point-to-point A connection topology where

two devices are linked directly to each other.
Point-to-Point Tunneling Protocol (PPTP)
PPTP is a Microsoft designed protocol that
lets companies extend their own corporate network through private tunnels over the public
Internet. Effectively, a corporation uses a wide
area network as a single large local area network. A company no longer needs to lease its
own lines for wide-area communication but
can securely use the public networks. This kind
of interconnection is known as a Virtual Private
Network.
portal Portal is a World Wide Web site that is
or proposes to be a major starting site for users
when they get connected to the Web or that
users tend to visit as an anchor site.
POST

See power on self-test (POST).

Post Office Protocol v 3 (POP3) POP3 is

used to accept and store e-mail and to allow
users to connect to their mailbox and access
their mail. SMTP is used to send mail to the
POP3 server.
PostScript A page-description language used
when printing high-quality text and graphics.
Desktop publishing or illustration programs
that create PostScript output can print on any
PostScript printer or imagesetter, because PostScript is hardware-independent. An interpreter
in the printer translates the PostScript commands into commands that the printer can
understand.
potentiometer

Copyright 2001 SYBEX, Inc., Alameda, CA

See variable resistor.

www.sybex.com

634

Glossary

power on self-test (POST) A set of diagnostic programs, loaded automatically from

ROM BIOS during startup, designed to ensure
that the major system components are present
and operating. If a problem is found, the POST
software writes an error message in the screen,
sometimes with a diagnostic code number indicating the type of fault located. These POST
tests execute before any attempt is made to load
the operating system.
power supply A part of the computer that
converts the power from a wall outlet into the
lower voltages, typically 5 to 12 volts DC,
required internally in the computer.
power surge A brief but sudden increase in
line voltage, often destructive, usually caused
by a nearby electrical appliance (such as a photocopier or elevator) or when power is reapplied after an outage.
power users A power user is someone who
either does administrative-level tasks on their
machine or needs to have additional access to
the system to do their work. The Power Users
group on a Windows 2000 Professional station
has abilities somewhere between normal users
and administrators.
preemptive multitasking A form of multitasking where the operating system executes an
application for a specific period of time,
according to its assigned priority and need. At
that time, it is preempted and another task is
given access to the CPU for its allocated time.
Although an application can give up control
before its time is up, such as during input/
output waits, no task is ever allowed to execute
for longer than its allotted time period.

Presentation layer The sixth of seven layers

of the International Standards Organizations
Open Systems Interconnection (ISO/OSI)
model for computer-to-computer communications. The Presentation layer defines the way
that data is formatted, presented, converted,
and encoded.
preventative maintenance The process of
performing various procedures on a computer
to prevent future data loss or system downtime.
Primary Domain Controller (PDC) The
main server in a Windows NT environment for
management and authentication. All changes
must be approved by the PDC before they can
be replicated out to the BDCs.
primary DOS partition In DOS, a division
of the hard disk that contains important operating system files. A DOS hard disk can be
divided into two partitions, or areas: the primary DOS partition and the extended DOS
partition. If you want to start your computer
from the hard disk, the disk must contain an
active primary DOS partition that includes the
three DOS system files: MSDOS.SYS, IO.SYS,
and COMMAND.COM. The primary DOS
partition on the first hard disk in the system is
referred to as drive C. Disk partitions are displayed, created, and changed using the FDISK
command.
print consumables Products that a printer
uses in the print process that must be replaced
occasionally. Examples include toner, ink, ribbons, and paper.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

printed-circuit board (PCB) Any flat board

made of plastic or fiberglass that contains chips
and other electronic components. Many PCBs
are multi-layer boards with several different
sets of copper traces connecting components
together.
printer control assembly Large circuit
board in the printer that converts signals from
the computer into signals for the various parts
in the laser printer.
printer ribbon A fabric strip that is impregnated with ink and wrapped around two spools
encased in a cartridge. This cartridge is used in
dot-matrix printers to provide the ink for the
print process.

635

product key Software piracy is a serious

problem in the industry, so many programs
include a product key that must be typed in for the
software to install properly. This key is then submitted if the user registers for technical support.
productivity tools Any of a number of applications users depend on to do job-related tasks.
Word processors and spreadsheets are common
examples, but most companies have additional
productivity tools as well.
Program Groups

See group icons.

Program Manager Group (GRP) Files Files

in the Windows 3.x directories that store information about which application icons are contained in which group icons.

printhead That part of a printer that creates

the printed image. In a dot-matrix printer, the
printhead contains the small pins that strike the
ribbon to create the image, and in an ink-jet
printer, the printhead contains the jets used to
create the ink droplets as well as the ink reservoirs. A laser printer creates images using an
electrophotographic method similar to that
found in photocopiers and does not have a
printhead.

Programmed Input/Output (PIO) One of

two transfer methods that are used in moving
data between the disk subsystem and memory.
PIO moves blocks of data from the disk subsystem directly in memory registers.

print media Another name for the mediums

being printed on. Examples include paper,
transparencies, and labels.

proprietary design A motherboard design

that is unique to a particular manufacturer and
is not licensed to other manufacturers.

priority field A field in a packet that determines what the priority of the packet is.

protected mode A processor operating

mode where every programs memory is protected from every other program so that if one
program crashes, it doesnt bring down the
other programs.

process An application or service running on

a server. This process will provide some type of
service to the server or to users.

Program Manager The primary interface to

Windows that allows you to organize and execute numerous programs by double-clicking an
icon in a single graphical window.

processor affinity Also called CPU affinity,

this is the process of selecting specific applications
or processes to run on a specific processor.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

636

Glossary

protocol In networking and communications, the specification that defines the procedures to follow when transmitting and
receiving data. Protocols define the format,
timing, sequence, and error-checking systems
used.
protocol stack In networking and communications, the several layers of software that
define the computer-to-computer or computerto-network protocol. The protocol stack on a
Novell NetWare system will be different from
that used on a Banyan VINES network or on a
Microsoft LAN Manager system.
proxy server A type of server that acts as a
firewall component to manage Internet traffic
to and from a local area network. A proxy
server can also provide Web page and document cacheing as well as other services,
including NAT.
PS/2 mouse interface A type of mouse
interface that uses a round, DIN-6 connector
that gets its name from the first computer it was
introduced on, the IBM PS/2.
Public Key Infrastructure (PKI) The PKI lets
users of an unsecure public network like the
Internet securely and privately exchange data
and money through the use of a public and a
private cryptographic key pair that is obtained
and shared through a trusted authority. The
PKI provides for a digital certificate that can
identify an individual or an organization and
directory services that can store and, when necessary, revoke the certificates.

Quality of Service (QoS) According to the

QoS Forum, A collective measure of the level of
service delivered to the customer. QoS can be
characterized by several basic performance criteria, including availability (low downtime),
error performance, response time and
throughput, lost calls or transmissions due to
network congestion, connection setup time,
and speed of fault detection and correction.
Service providers may guarantee a particular
level of QoS (defined by a service level agreement or SLA) to their subscribers.
QoS-enabled hardware and software solutions sort and classify IP packet requests into
different traffic classes and allocate the proper
resources to direct traffic based on various criteria including application type, user or application ID, source or destination IP address,
time of day, and other user-specified variables.
QSOP (Quad Small Outline Package) A
type of IC package that has all leads soldered
directly to the circuit board. Also called a surface mount chip.
Quick-and-Dirty Disk Operating System
(QDOS) Created by Tim Patterson of Seattle
Computer Products, QDOS was the basis of
MS-DOS. QDOS was purchased by Microsoft
and renamed MS-DOS.
raceway

Track for running cables over walls.

rack A device designed to hold multiple

devices like servers, UPSs, backup devices, etc.,
in one lockable cabinet.

puck The proper name for the mouse-like

device used with drawing tablets.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

radio frequency interference (RFI) Many

electronic devices, including computers and
peripherals, can interfere with other signals in the
radio-frequency range by producing electromagnetic radiation; this is normally regulated by government agencies in each country.
RAID 0 RAID 0 is known as data striping
writing data across several disks. This technique has no redundancy of data. It offers the
best performance but no fault-tolerance.
RAID 0+1 Pure RAID 1 suffers from the same
problems as pure RAID 0: data is written
sequentially across the volumes, potentially
making one drive busy while the others on that
side of the mirror are idle. You can avoid this
problem by striping within your mirrored volumes, just like you striped within the volumes
of your RAID 0 set.
Theoretically, RAID 1 mirrors physical
devices on a one-for-one basis. In reality, you'll
want to build big logical devices using RAID 0
sets and then mirror them for redundancy. This
configuration is known as RAID 0+1, since it
combines the concatenation features of RAID 0
with the mirroring features of RAID 1.
RAID 1 This type is also known as disk mirroring and consists of at least two drives that
duplicate the storage of data. There is no
striping. Read performance is improved since
either disk can be read at the same time. Write
performance is the same as for single disk
storage. RAID-1 provides the best performance
and the best fault-tolerance in a multi-user
system.

637

RAID 5 This type includes a rotating parity

array, thus addressing the write limitation in
RAID 4. Thus, all read and write operations
can be overlapped. RAID 5 stores parity information but not redundant data (although
parity information can be used to reconstruct
data). RAID 5 requires at least three and usually five disks for the array. It's best for multiuser systems in which performance is not critical or which do few write operations.
RAID cache Memory addressable by a RAID
host bus adapter to temporarily store information that is going to be written to the disk subsystem.
RAM Acronym for random access memory.
The main system memory in a computer, used
for the operating system, application programs,
and data.
RAM disk An area of memory managed by a
special device driver and used as a simulated disk.
Anything stored on a RAM disk will be erased
when the computer is turned off; therefore, the
contents must be saved onto a real disk.
Rambus (RDRAM)

See Direct Rambus.

Rambus Inline Memory Modules

(RIMMs) A type of memory module that uses
Rambus memory. See Direct Rambus.
random access memory

See RAM.

rasterizing The process of converting signals

from the computer into signals for the various
assemblies in the laser printer.
read-only memory
memory).

Copyright 2001 SYBEX, Inc., Alameda, CA

See ROM (read-only

www.sybex.com

638

Glossary

read/write head That part of a floppy- or

hard-disk system that reads and writes data to
and from a magnetic disk.

REM statement A command placed in the

beginning of a line in a DOS batch file to prevent that line from executing.

real mode A processor operating mode

whereby a processor emulates an 8086 processor.

Remote Access Server (RAS) A server dedicated to providing remote users access to the
LAN. It could be considered a modem farm.

Reduced Instruction Set Code (RISC)

Microprocessor design that uses a relatively
small set of instructions to perform efficient
and rapid processing.

Remote Authentication Dial In User Service

(RADIUS) An Internet protocol in which
authentication servers can provide authorization and authentication information to a network server to which the user is attempting to
connect.

Redundant Array of Independent Disks

(RAID) A data storage system where data and
information needed for error correction are
stored over two or more hard disks to improve
reliability and performance.
reference disk A special disk that is bootable
and contains a program that is able to send special commands to MCA bus devices to configure their parameters.
refresh rate In a monitor, the rate at which
the phosphors that create the image on the
screen are recharged.
registered memory A specially designed
dynamic RAM that supports access to sequential
memory locations with a reduced cycle time.
registration roller
roller.

See paper registration

Registry The Registry is used in Windows

9x, NT, and 2000 to store configuration information about the machine. This includes information about both individual user settings and
global system settings.

removable mass storage Any highcapacity storage device inserted into a drive for
reading and writing, then removed for storage
and safekeeping.
removable media Any storage media that
can be removed from the system.
repeater In networking, a simple hardware
device that moves all packets from one local
area network segment to another.
reserved memory In DOS, a term used to
describe that area of memory between 640K
and 1MB, also known as upper memory.
Reserved memory is used by DOS to store
system and video information.
Reset Server A command that can be used
on a Novell Server to stop the network operating system, exit to DOS, and perform a warm
boot.
resilient server links Links between servers
that can bounce back after an interruption.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

resistor pack A combination of multiple

resistors in a single package. Often used for terminating SCSI busses.
resistor An electronic device used to resist
the flow of current in an electrical circuit. See
also fixed resistor and variable resistor.
resource Anything on a network that clients
might want to access or use.
Restart Server A command to be used on a
NetWare Server to stop and restart the network
operating system.
restore The process of getting data from a
backup restored to the computer it originally
came from.
Reverse Proxy Server A Proxy Server that
serves users coming in from the public network,
as opposed to a Proxy Servers usual function
of serving users on a private network.
RFI

See radio frequency interference (RFI).

rheostat

See variable resistor.

ribbon cartridge The container that holds

the printer ribbon.
ring topology Type of physical topology in
which each computer connects to two other
computers, joining them in a circle and creating
a unidirectional path where messages move
from workstation to workstation. Each entity
participating in the ring reads a message, regenerates it, and then hands it to its neighbor.
RJ-11/RJ-45 A commonly used modular
telephone connector. RJ-11 is a four- or six-pin
connector used in most connections destined

639

for voice use; it is the connector used on phone

cords. RJ-45 is the eight-pin connector used for
data transmission over twisted-pair wiring and
can be used for networking; RJ-45 is the connector used on 10Base-T Ethernet cables.
RLL encoding
encoding.

See run-length limited (RLL)

ROM (read-only memory) A type of computer memory that retains its data permanently, even when power is removed. Once the
data is written to this type of memory, it cannot
be changed.
root directory In a hierarchical directory
structure, the directory from which all other
directories must branch. The root directory is
created by the FORMAT command and can
contain files as well as other directories. This
directory cannot be deleted.
router In networking, an intelligent connecting device that can send packets to the correct local area network segment to take them to
their destination. Routers link local area network segments at the network layer of the
International Standards Organizations Open
Systems Interconnect (ISO/OSI) model for
computer-to-computer communications.
Routing Information Protocol (RIP) RIP is
a protocol that is used to route packets within
larger networks. It is used instead of other protocols like the Open Shortest Path First (OSPF).
RIP is an older routing protocol that is installed
in many of today's corporate networks. RIP is
designated by the Internet Engineering Task
Force (IETF) as one of several Interior Gateway
Protocols.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

640

Glossary

RS-232-C In asynchronous transmissions, a

recommended standard interface established
by the Electrical Industries Association. The
standard defines the specific lines, timing, and
signal characteristics used between the computer and the peripheral device and uses a 25pin or 9-pin DB connector. RS-232-C is used
for serial communications between a computer
and a peripheral such as a printer, modem, digitizing tablet, or mouse.
RS-232 cables

See serial cables.

RS-422/423/449 In asynchronous transmissions, a recommended standard interface established by the Electrical Industries Association for
distances greater than 50 feet but less than 1000
feet. The standard defines the specific lines,
timing, and signal characteristics used between
the computer and the peripheral device.
RTS Abbreviation for request to send. A
hardware signal defined by the RS-232-C standard to request permission to transmit.
run-length limited (RLL) encoding An efficient method of storing information on a hard
disk that effectively doubles the storage
capacity of a disk when compared to older, less
efficient methods such as modified frequency
modulation encoding (MFM).
Safe Mode A Windows 9x operating mode
that only loads a basic set of drivers and a basic
screen resolution. It can be activated using the
F8 key at boot time.
scanner An optical device used to digitize
images such as line art or photographs, so that
they can be merged with text by a page-layout

or desktop publishing program or incorporated

into a CAD drawing.
scope When used with Dynamic Host Configuration Protocol, a scope is a range of
Internet Protocol (IP) addresses that can be
leased to client hosts.
screen saver Program originally designed to
prevent damage to a computer monitor from
being left on too long. These programs usually
include moving graphics so that no one pixel is
left on all the time. Screen savers detect computer inactivity and activate after a certain
period.
SCSI Acronym for small computer system
interface. A high-speed, system-level parallel
interface defined by the ANSI X3T9.2 committee. SCSI is used to connect a personal computer to several peripheral devices using just
one port. Devices connected in this way are said
to be daisy-chained together, and each
device must have a unique identifier or priority
number.
SCSI adapter Device that is used to manage
all the devices on the SCSI bus as well as to send
and retrieve data from the devices.
SCSI address
SCSI device.

A unique address given to each

SCSI bus Another name for the SCSI interface and communications protocol.
SCSI chain All the devices connected to a
single SCSI adapter.
SCSI Fast-20 Implementation of SCSI technology that transfers data at 20Mbps.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

SCSI Fast-40 Implementation of SCSI technology that transfers data at 40Mbps.

SCSI Narrow This is the first implementation of SCSI technology. All data was transferred asynchronously and the SCSI bus was 8
bits wide, so this was considered SCSI Narrow.
SCSI-2 SCSI communication in SCSI-2 was
shifted from asynchronous to synchronous,
and the size of the SCSI bus was increased to 16
bits. This increased the data transfer rate to
between 5Mbps and 20Mbps and also necessitated a change in the type of cabling and connectors.
SCSI terminator The SCSI interface must be
correctly terminated to prevent signals echoing
on the bus. Many SCSI devices have built-in
terminators that engage when they are needed.
With some older SCSI devices, you have to add
an external SCSI terminator that plugs into the
devices SCSI connector.
secondary zone In DNS, a secondary zone is
a subset of the primary DNS namespace. For
example, your primary zone DNS name space
might be ABCDEFG.COM. Because your company has multiple locations, you may break
these zones up into secondary zones like
Minneapolis.ABCDEFG.COM and
Orlando.ABCDEFG.COM.
sector The smallest unit of storage on a disk,
usually 512 bytes. Sectors are grouped together
into clusters.

641

Often used as a performance gauge of an individual drive. The major part of a hard disks
access time is actually seek time.
self-healing driver Some of the more
advanced NIC cards will constantly test the
communication chain to make sure things are
working as planned. If the NIC comes up with
driver problems, the card can attempt to repair
or replace the driver. This is referred to as a
self-healing driver.
Self-Monitoring Analysis and Report Technology (SMART) In IDE hard disk technology, the hard disk will continuously
examine its level of operation. If the disk finds
that there are problems, it will report the
problem. The reports will occur on system boot
as part of the Power On Self-Test.
semiconductors Any material that,
depending on some condition, is either a conductor or non-conductor.
Sequenced Packet Exchange (SPX) SPX is
used primarily in Novell NetWare networks as
the reliable, connection-oriented protocol.
Serial ATA At the time of this writing, Serial
ATA was the latest method of communication
between the system bus and the ATA devices.
Serial communications, as opposed to parallel
communications, provides much faster
throughput and performance.
serial cables Cables used for serial communications. See serial communications.

seek time Time it takes the actuator arm to

move from rest position to active position for
the read/write head to access the information.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

642

Glossary

serial communications The transmission of

information from computer to computer or
from computer to a peripheral, one bit at a
time. Serial communications can be synchronous and controlled by a clock or asynchronous and coordinated by start and stop bits
embedded in the data stream.
serial mouse A mouse that attaches directly
to one of the computers serial ports.
serial port A computer input/output port
that supports serial communications in which
information is processed one bit at a time. RS232-C is a common serial protocol used by
computers when communicating with
modems, printers, mice, and other peripherals.
serial printer A printer that attaches to one
of the computers serial ports.
server In networking, any computer that
makes access to files, printing, communications, or other services available to users of the
network. In large networks, a server may run a
special network operating system; in smaller
installations, a server may run a personal computer operating system.
service A service is any program that runs in
the background on a computer and performs
some sort of task for that computer or other
machines on the network.

service log The service log is a written record

of any service performed on a server or on a
workstation. This service can be as simple as
updating a driver to as complex as reconfiguring
the entire system. By providing a service log, the
technician knows what has been done to the computer recently, and can make troubleshooting
decisions based on that information.
Session layer The fifth of seven layers of the
International Standards Organizations Open
Systems Interconnection (ISO/OSI) model for
computer-to-computer communications. The
Session layer coordinates communications and
maintains the session for as long as it is needed,
performing security, logging, and administrative functions.
share name The share name is used to identify a network access point. Share names can be
the same as the directory they are sharing or
they can be different.
shell Every operating system needs to have
some sort of interface that allows users to navigate the system. The shell is the program that
controls how this interface works. For MSDOS, the Windows Program Manager was its
most popular shell. For Windows 9x and 2000,
Explorer (explorer.exe) is the standard shell
program.

Service Advertising Protocol (SAP) In a

Novell NetWare network, SAP is used to advertise services like print servers or file servers.

shell scripting Writing a script to be executed by the command interpreter of the operating system. This is common in Linux and
UNIX environments.

Service Layer Layer of the I2O model that

provides the actual services. See I2O.

shielded twisted-pair
twisted-pair).

Copyright 2001 SYBEX, Inc., Alameda, CA

See STP (shield

www.sybex.com

Glossary

Shut Down A procedure used on all major

NOSs to stop the server from providing services
and to take it offline. The method for shutting
down the server will depend on the operating
system.
signal ground Wiring specification for
cables, making sure a cable is grounded.
Simple Mail Transport Protocol
(SMTP) SMTP is used to send mail from a
client to an e-mail server. SMTP servers do not
store mail for users to pick up; they simply send
the mail out, and another server (such as a
POP3 server) is used to store incoming mail.
Simple Network Management Protocol
(SNMP) SNMP is used to manage and track
the network. It consists of an SNMP Manager
and an SNMP Agent. The Agent gathers statistics on hardware or software events and saves
those statistics until the SNMP Manager asks
for them. The SNMP Agent and Manager can
also be configured to provide alerts when there
are problems detected on the network.
Single Inline Memory Module
(SIMM) Individual RAM chips are soldered
or surface mounted onto small narrow circuit
boards called carrier modules, which can be
plugged into sockets on the motherboard.
These carrier modules are simple to install and
occupy less space than conventional memory
modules.
Single Inline Package (SIP) A type of semiconductor package where the package has a
single row of connector pins on one side only.

643

single-purpose server A server that is dedicated to one purpose (e.g., a file server or a
printer server).
site license A software license that is valid
for all installations at a single site.
slave

See slave drive.

slave drive The secondary drive in a IDE

master/slave disk configuration.
small computer system interface
SCSI.

See

Small Outline DIMM A memory module

designed for use in smaller devices like a laptop
or, in some cases, a printer.
SmartCard A method of network authentication that involves a small, credit cardlike
device that contains information about the
user. This device can store an authentication
key. SmartCards are usually configured to use
another form of authentication like a password
or a biometric scan.
SNMP Agent

See SNMP.

SNMP Manager

See SNMP.

socket services Part of the software support

needed for PCMCIA hardware devices in a portable computer, controlling the interface to the
hardware. Socket services is the lowest layer in
the software that manages PCMCIA cards. It
provides a BIOS-level software interface to the
hardware, effectively hiding the specific details
from higher levels of software. Socket services
also detect when you insert or remove a
PCMCIA card and identify the type of card it is.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

644

Glossary

soft page fault If a page fault occurs and the

data is still in a free frame, rather than being in
RAM, it is a soft page fault. See page fault.
software An application program or an
operating system that a computer can execute.
Software is a broad term that can imply one or
many programs, and it can also refer to applications that may actually consist of more than
one program.
software driver Software that acts as the
liaison between a piece of hardware and the
operating system and allows the use of a component.
solenoid An electromechanical device that,
when activated, produces an instant push or
pull force.
source All computer programsoperating
system or applicationare nothing but a collection of program code. This is the source code
or source that defines what a program is and
how it works. The open source movement is
involved with allowing you to see and even
modify this code.
spin speed An indication of how fast the
platters on a fixed disk are spinning.
spindle The rod that platters are mounted to
on in a hard disk drive.
SRAM

See static RAM (SRAM).

ST506 interface A popular hard-disk interface standard developed by Seagate Technologies, first used in IBMs PC/XT computer and
still popular today, with disk capacities smaller
than about 40MB. ST506 has a relatively slow
data transfer rate of 5 megabits per second.

stack Another name for the memory map, or

the way memory is laid out.
standard peripheral power connector
Type of connector used to power various
internal drives. Also called a Molex connector.
standby power supply (SBS) A power
supply designed to provide power to computer
equipment in the event of a complete power
outage. A standby power supply could be
something like a generator.
star network A network topology in the
form of a star. At the center of the star is a
wiring hub or concentrator, and the nodes or
workstations are arranged around the central
point representing the points of the star.
start bit In asynchronous transmissions, a
start bit is transmitted to indicate the beginning
of a new data word.
Start menu As the main focus of the Windows 9x/NT/2000 user interface, the Start
menu allows program shortcuts to be placed
for easy and organized access.
static RAM (SRAM) A type of computer
memory that retains its contents as long as
power is supplied. It does not need constant
refreshment like dynamic RAM chips.
static-charge eliminator strip The device in
EP process printers that drains the static charge
from the paper after the toner has been transferred to the paper.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

steering IP address In a NIC Load Balancing solution, a steering IP address is the

address that lets the Load Balancing software
intercept the hand-shaking process and rotate
the NICs MAC address between all the NICs
in a round-robin fashion. That way, all the of
the NICs in the group will have an equal
number of clients.
stepper motor A very precise motor that can
move in very small increments. Often used in
printers.
stop bit(s) In asynchronous transmissions,
stop bits are transmitted to indicate the end of
the current data word. Depending on the convention in use, one or two stop bits are used.
Storage Area Network (SAN) A series of
network devices that are designed around a set
of large storage arrays. In a storage area network, multiple servers may be storing data or
accessing information from a single group of
hard disks.
Streaming Single-Instruction, MultipleData Extensions A category of parallel processor computer architecture where one
instructor processor gets instructions and sends
orders to other processors.
STP (shield twisted-pair) Cabling that has a
braided foil shield around the twisted pairs of
wire to decrease electrical interference.
stylus A pen-like pointing device used in penbased systems and personal digital assistants.

645

subnet mask The subnet mask is a required

part of any TCP/IP configuration, and it is used
to define which addresses are local and which
are on remote networks.
superscalar

See parallel processing.

SuperVGA (SVGA) An enhancement to the

Video Graphics Array (VGA) video standard
defined by the Video Electronics Standards
Association (VESA).
surface mount See Quad Small Outline
Package (QSOP).
surge suppressor Also known as a surge
protector. A regulating device placed between
the computer and the AC line connection that
protects the computer system from power
surges.
SVGA

See SuperVGA (SVGA).

swap file On a hard disk, a file used to store

parts of running programs that have been
swapped out of memory temporarily to make
room for other running programs. A swap file
may be permanent, always occupying the same
amount of hard disk space even though the
application that created it may not be running,
or is temporary, only created as and when
needed.
switch ports As part of the network communication process, all communication may go
through a hub or a switch. A switch port is
where the cable from the host plugs into the
switch.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

646

Glossary

Symmetrical Multiprocessing (SMP)

Having two or more processors working
together in a single computer.
synchronization The timing of separate elements or events to occur simultaneously.
1. In a multimedia presentation, synchronization ensures that the audio and video components
are timed correctly, so they actually make sense.
2. In computer-to-computer communications, the hardware and software must be synchronized so that file transfers can take place.
3. The process of updating files on both a
portable computer and a desktop system so
that they both have the latest versions is also
known as synchronization.
synchronous clock encoding technique
Encoding managed by a mutually agreed
upon timing source.
synchronous DRAM A type of DRAM
memory module that uses memory chips synchronized to the speed of the processor.
synchronous transmission In communications, a transmission method that uses a clock
signal to regulate data flow. Synchronous
transmissions do not use start and stop bits.
syntax Syntax is a term used to describe the
proper way of forming a text command for
entry into the computer. Many commands have
a number of different options, each of which
requires a particular format.
system attribute Attribute of DOS that is
used to tell the OS that this file is needed by the
OS and should not be deleted. Marks a file as
part of the operating system and will also protect the file from deletion.

system board The sturdy sheet or board to

which all other components on the computer
are attached. These components consist of the
CPU, underlying circuitry, expansion slots,
video components, and RAM slots, just to
name a few. Also known as a logic board,
motherboard, or planar board.
system bus The series of communication
channels on the system board.
system disk A disk that contains all the files
necessary to boot and start the operating
system. In most computers, the hard disk is the
system disk; indeed, many modern operating
systems are too large to run from floppy disk.
SYSTEM.INI In Microsoft Windows, an initialization file that contains information on
your hardware and the internal Windows operating environment.
System Network Architecture (SNA) IBM
protocols for communication with a mainframe computer.
system resources On a Windows 3.x or 95/
98 machine, the system resources represent
those components of the PC that are being used
(memory, CPU, etc.).
system software The programs that make
up the operating system, along with the associated utility programs, as distinct from an application program.
tabs On many windows you will find that, to
save space, a single window will have many
tabs, each of which can be selected to display
particular information.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

tape cartridge A self-contained tape storage

module, containing tape much like that in a
video cassette. Tape cartridges are primarily
used to back up hard disk systems.
tape drive Removable media drive that uses
a tape cartridge that has a long polyester ribbon
coated with magnetic oxide and wrapped
around two spools with a read/write head in
between.
target Another name for the backup media,
it is the destination for the data being backed
up. It is usually a tape drive or other backup
device.
taskbar The area of the Windows 9x/NT/
2000 interface which includes the Start button
and the System Tray, as well as icons for any
open programs.
TCP/IP Acronym for Transmission Control
Protocol/Internet Protocol. A set of computerto-computer communications protocols that
encompass media access, packet transport, session communications, file transfer, e-mail, and
terminal emulation. TCP/IP is supported by a
very large number of hardware and software
vendors and is available on many different
computers from PCs to mainframes.

647

temporary swap file A swap file that is created every time it is needed. A temporary swap
file will not consist of a single large area of contiguous hard disk space, but may consist of several discontinuous pieces of space. By its very
nature, a temporary swap file does not occupy
valuable hard disk space if the application that
created it is not running. In a permanent swap
file the hard disk space is always reserved and is
therefore unavailable to any other application
program.
terabyte A measure used for high capacity
data storage. One terabyte equals 240 bytes,
although it is usually associated with 1 trillion
bytes.
terminal A monitor and keyboard attached
to a computer (usually a mainframe), used for
data entry and display. Unlike a personal computer, a terminal does not have its own central
processing unit or hard disk.
terminal emulation mode Terminal emulation mode is a method of putting a PC into a
configuration where it acts like a computer terminal without any processing power. A terminal will just provide commands to a
mainframe and all commands will be processed
at the mainframe.

Telecommunications Industry Association

(TIA) TIA is the organization that is recognized in the United States as the body that
develops telephony standards.

Terminate and Stay Resident (TSR) A

DOS program that stays loaded in memory, even
when it is not actually running, so that you can
invoke it very quickly to perform a specific task.

Telnet A member of the TCP/IP protocol

suite that allows a user to establish a session
with a remote host and take over that host.

termination The act of terminating a SCSI

chain. See terminator.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

648

Glossary

terminator A device attached to the last

peripheral in a series or the last node on a network. A resistor is placed at both ends of a coax
Ethernet cable to prevent signals from
reflecting and interfering with the transmission.

thrashing A slang term for the condition that

occurs when Windows must constantly swap
data between memory and hard disk. The hard
disk spins continuously during this and makes a
lot of noise.

text mode A video display mode for a video

card that allows it to only display text. When
running DOS programs, a video card is in text
mode.
The VL bus is a 32-bit bus, running at either
33MHz or 40MHz. The maximum throughput
is 133 megabytes/second at 33MHz, or 148
megabytes/second at 40MHz. The most
common VL bus adapters are video adapters,
hard-disk controllers, and network interface
cards.

thread A single process that can be allocated to

perform work at the processor level. Each thread
has the ability to perform a particular task.

thermal printer A nonimpact printer that

uses a thermal printhead and specially treated
paper to create an image.

Token Ring network A local area network

with a ring structure that uses token-passing to
regulate traffic on the network and avoid collisions. On a Token Ring network, the controlling computer generates a token that
controls the right to transmit. This token is
continuously passed from one node to the next
around the network. When a node has information to transmit, it captures the token, sets its
status to busy, and adds the message and the
destination address. All other nodes continuously read the token to determine if they are the
recipient of a message; if they are, they collect
the token, extract the message, and return the
token to the sender. The sender then removes
the message and sets the token status to free,
indicating that it can be used by the next node
in sequence.

thick Ethernet Connecting coaxial cable

used on an Ethernet network. The cable is 1 cm
(approximately 0.4") thick and can be used to
connect network nodes up to a distance of
approximately 3300 feet. Thick Ethernet is primarily used for facility-wide installations. Also
known as 10Base5.
thin Ethernet Connecting coaxial cable used
on an Ethernet network. The cable is 5 mm
(approximately 0.2) thick, and can be used
to connect network nodes up to a distance of
approximately 1000 feet. Thin Ethernet is primarily used for office installations. Also known
as 10Base2.

token passing A media access method that

gives every NIC equal access to the cable. The
token is a special packet of data that is passed
from computer to computer. Any computer
that wants to transmit has to wait until it has
the token, at which point it can add its own
data to the token and send it on.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

tolerance band Found on a fixed resistor,

this colored band indicates how well the
resistor holds to its rated value.
toner Black carbon substance mixed with
polyester resins and iron oxide particles.
During the EP printing process, toner is first
attracted to areas that have been exposed to the
laser in laser printers and is later deposited and
melted onto the print medium.
toner cartridge The replaceable cartridge in
a laser printer or photocopier that contains the
electrically charged ink to be fused to the paper
during printing.
topology A way of laying out a network.
Can describe either the logical or physical
layout.
total cost of ownership (TCO) A method of
determining the actual cost of a server or a
device. The TCO goes beyond the invoice price
to include administration costs, maintenance
costs, etc.
touch screen A special monitor that lets the
user make choices by touching icons or graphical buttons on the screen.
Tracert Used to trace the path of a packet
across a TCP/IP network.
trackball An input device used for pointing,
designed as an alternative to the mouse.
tracks The concentric circle unit of hard disk
division. A disk platter is divided into these
concentric circles.

649

transfer corona assembly The part of an EP

process printer that is responsible for transferring the developed image from the EP drum to
the paper.
transfer step The step in the EP print process
where the developed toner image on the EP
drum is transferred to the print media using the
transfer corona.
transistor Abbreviation for transfer resistor.
A semiconductor component that acts like a
switch, controlling the flow of an electric current. A small voltage applied at one pole controls a larger voltage on the other poles.
Transistors are incorporated into modern
microprocessors by the million.
Transmission Control Protocol/Internet
Protocol See TCP/IP.
Transmission Protocols Protocols designed
to transmit packets across computer networks.
Transport layer The fourth of seven layers of
the International Organization for Standardizations Open Systems Interconnection (ISO/
OSI) model for computer-to-computer communications. The Transport layer defines protocols for message structure and supervises the
validity of the transmission by performing
some error checking.
TSR

See Terminate and Stay Resident (TSR).

twisted-pair cable Cable that comprises two

insulated wires twisted together at six twists
per inch. In twisted-pair cable, one wire carries
the signal and the other is grounded. Telephone
wire installed in modern buildings is often
twisted-pair wiring.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

650

Glossary

UART Acronym for Universal Asynchronous

Receiver/Transmitter. An electronic module
that combines the transmitting and receiving
circuitry needed for asynchronous transmission
over a serial line. Asynchronous transmissions
use start and stop bits encoded in the data
stream to coordinate communications rather
than the clock pulse found in synchronous
transmissions.
Ultra 2 SCSI A SCSI specification that allows
for transfer of data at 40Mbps across a 16-bit bus.
Ultra DMA 33 ATA specifications that allow
a data transfer rate of 33Mbps.
Ultra DMA IDE Also known as ATA version
4 (ATA-4), it can transfer data at 33Mbps, so it
is also commonly seen in motherboard specifications as Ultra DMA/33, Ultra 66, or UDMA.
Ultra SCSI

See SCSI-2.

unbuffered memory The process of not

storing information in memory. Instead, the
information is processed as it is received.
uninstall To remove a program from a computer. This generally involves removing its configuration information from the Registry, its
icons from the Start menu, and its program
code from the file system.
Uninterruptible Power Supply (UPS)
Backup power supply for the computer room.
A UPS will allow server operations to continue
even if power is lost. Generally, a UPS can maintain operations for between 15 and 30 minutes.
UPS systems can also provide power conditioning
services to computers, removing power spikes
and also leveling out dips in power.

uniprocessor (UP) A system with a single

processor or CPU.
Universal Serial Bus

See USB.

UNIX A 32-bit, multiuser, multitasking,

portable operating system.
Unshielded Twisted Pair (UTP)
upper memory area
area.

See UTP.

See reserved memory

upper memory block (UMB) Free areas of

memory that can be used for loading drivers
and programs into the upper memory area.
USB Acronym for Universal Serial Bus. A
technology used to connect peripheral devices
to a computer. Each USB channel will support
127 devices and has a total transfer rate of up
to 12MBps.
USER.EXE Windows core component that
allows a user to interact with Windows. It is the
component responsible for interpreting keystrokes and mouse movements and sending the
appropriate commands to the other core components.
username In order to identify themselves on
the network, each user must provide two credentialsa username and a password. The
username says, This is who I am, and the
password says, And heres proof! Each username must be unique on the network and is
generally used by only one person.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

user profiles In order to allow each user to

customize their Windows experience, user profiles save a particular users desktop appearance and preferences so that when they log on,
they will always have there own desktop, even
if they share the machine with others.
user thread User level threads implement a
priority scheme and queue-management system
distinct from the kernel thread priorities and
per-processor dispatch queues that exist in the
kernel.
utility program A small program or set of
small programs that support the operating
system by providing additional services that the
operating system does not provide.
UTP Acronym for Unshielded Twisted Pair.
A type of unshielded network cable that contains multiple conductors in pairs that are
twisted around each other.
vaccine An application program that
removes and destroys a computer virus. The
people who unleash computer viruses are often
very accomplished programmers, and they are
constantly creating new and novel ways of
causing damage to a system. The antivirus and
vaccine programmers do the best they can to
catch up, but they must always lag behind to
some extent.
vacuum tube Electronic component that is a
glorified switch. A small voltage at one pole
switches a larger voltage at the other poles on
or off.

651

variable resistor A resistor that does not

have a fixed value. Typically the value is
changed using a knob or slider.
version Each time that computer software is
modified, new features are added and old problems are, hopefully, fixed. To tell these modified programs apart, computer programmers
use versions. These are incremented by one
digit (for example, from 1.0 to 2.0) for major
revisions, or by a tenth of a digit (for example,
from 2.0 to 2.1) for minor modifications.
Higher version numbers mean newer versions.
VGA Acronym for Video Graphics Array. A
video adapter. VGA supports previous graphics
standards, and provides several different
graphics resolutions, including 640 pixels horizontally by 480 pixels vertically. A maximum
of 256 colors can be displayed simultaneously,
chosen from a palette of 262,114 colors.
Because the VGA standard requires an analog
display, it is capable of resolving a continuous
range of gray shades or colors. In contrast, a
digital display can only resolve a finite range of
shades or colors.
video adapter An expansion board that
plugs into the expansion bus in a DOS computer and provides for text and graphics output
to the monitor. The adapter converts the text
and graphic signals into several instructions for
the display that tell it how to draw the graphic.
Video Graphics Array

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

See VGA.

652

Glossary

video RAM (VRAM) Special-purpose RAM

with two data paths for access, rather than just
one as in conventional RAM. These two paths
let a VRAM board manage two functions at
oncerefreshing the display and communicating with the processor. VRAM doesnt
require the system to complete one function
before starting the other, so it allows faster
operation for the whole video system.
virtual circuit A connection between two
computers that gives the computers what
appears to be a direct link, but may actually
involve routing data over a long path.
virtual LAN A local area network that consists of several groups of hosts that are physically located on different network segments but
that communicate as though they were on the
same wire.
virtual memory A memory-management
technique that allows information in physical
memory to be swapped out to a hard disk. This
technique provides application programs with
more memory space than is actually available in
the computer. True virtual-memory management
requires specialized hardware in the processor for
the operating system to use; it is not just a question of writing information out to a swap file on
the hard disk at the application level.
virtual NIC A NIC that can be referenced, but
does not physically exist.
Virtual Private Network (VPN) Using a set
of protocols to establish a virtual circuit
between a two hosts over a public network. For
example, you dial in to your Internet Service
Provider and establish a virtual circuit to a

router on your enterprise network, giving you

all the services available on the private network. This is referred to as a Virtual Private
Network.
virus A program intended to damage your
computer system without your knowledge or
permission. A virus may attach itself to another
program or to the partition table or the boot
track on your hard disk. When a certain event
occurs, a date passes, or a specific program executes, the virus is triggered into action. Not all
viruses are harmful; some are just annoying.
VL bus Also known as VL local bus. Abbreviation for the VESA local bus, a bus architecture introduced by the Video Electronics
Standards Association (VESA), in which up to
three adapter slots are built into the motherboard. The VL bus allows for bus mastering.
VLSI (Very Large Scale Integration) Technology used by chip manufacturers to integrate
the functions of several small chips into one chip.
volts
VRAM

Unit of electrical potential.

See video RAM (VRAM).

wait state A clock cycle during which no

instructions are executed because the processor
is waiting for data from a device or from
memory.
WAN (wide area network) Network that
expands LANs to include networks outside of
the local environment and also to distribute
resources across distances.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

warm boot Refers to pressing Control+Alt+Delete to reboot the computer. This

type of booting doesnt require the computer to
perform all of the hardware and memory
checks that a cold boot does.
warm start

See warm boot.

Warp Server The name of the IBM OS/2

network operating system.
Web-based Distributed Authoring and Versioning (WebDAV) A new utility provided
with NetWare 5.1.
well known ports Ports used by protocols in
the TCP/IP protocol suite. A listing of these
ports can be found in a file called SERVICES in
a directory called \ETC. The location of the
ETC directory varies, depending on the operating system.
what-you-see-is-what-you-get (wysiwyg)
Printing and desktop publishing technology
where what appears on the computer monitor
is what can be printed.
wide area network See WAN (wide area
network).
window In a graphical user interface, a rectangular portion of the screen that acts as a
viewing area for application programs. Windows can be tiled or cascaded and can be individually moved and sized on the screen. Some
programs can open multiple document windows inside their application window to display several word processing or spreadsheet
data files at the same time.
Windows 95 Windows 95 is a 32-bit, multitasking, multithreaded operating system
capable of running DOS, Windows 3.1, and

653

Windows 95 applications; supports Plug and

Play (on the appropriate hardware); and adds
an enhanced FAT file system in the Virtual FAT
which allows long filenames of up to 255 characters while also supporting the DOS 8.3 filenaming conventions.
Windows 98 The home PC operating system
released by Microsoft as the successor to their
popular Windows 95 operating system. Basically the same as Windows 95, it offers a few
improvements. For example, Windows 98
improves upon the basic look and feel of Windows 95 with a new browser-like interface. It
also contains bug fixes and can support two
monitors simultaneously. In addition to new
interface features, it includes support for new
hardware, including Universal Serial Bus
devices.
Windows 2000 The newest Windows operating system, which incorporates the look and
feel of Windows 9x with the power of
Windows NT.
Windows Desktop

See desktop.

Windows Installer A new method Microsoft

is using to allow users to customize their application installations more easily. The Windows
Installer also makes it easier for users to install
approved software on secured workstations
and can automatically repair damaged installs.
Windows Internet Name Service
(WINS) WINS provides a database for the
storage and retrieval of NetBIOS computer
names. Each client must register with the WINS
server to be able to be added to and query the
database.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

654

Glossary

Windows NT A 32-bit multitasking portable

operating system developed by Microsoft. Windows NT is designed as a portable operating
system, and initial versions run on Intel 80386
(or later) processors and RISC processors, such
as the MIPS R4000 and the DEC Alpha. Windows NT contains the graphical user interface
from Windows 3.1, and can run Windows 3.1
and DOS applications as well as OS/2 16-bit
character-based applications and new 32-bit
programs specifically developed for Windows
NT. Multitasking under Windows NT is preemptive, and applications can execute multiple
threads. Security is built into the operating
system at the U.S. Government-approved C2
security level. Windows NT supports the DOS
FAT file system, the OS/2 HPFS, installable file
systems such as CD-ROM systems, and a
native file system called NTFS. Windows NT
also supports multiprocessing, OLE, and peerto-peer networking.
Windows Program Manager Windows 3.x
file that contains all of the program icons,
group icons, and menus used for organizing,
starting, and running programs.
WIN.INI File that contains Windows environmental settings that control the environments
general function and appearance.
WINIPCFG In Windows 9x, this is the utility
that allows you to view your current TCP/IP
configuration. It also allows a user to request a
new IP configuration from a DHCP server.
wiring diagram A graphical representation
of the way the computer network is wired. This
can include things like the speed of the connections, the types of cabling, the place of terminators (if required), and any other information

necessary to improve or simplify the troubleshooting process.

wizard Wizards are pre-programmed utilities
that walk the user through a particular task.
Each wizard generally includes a number of different pages, each of which allows you to enter
information or choose particular options. At
the finish of the wizard, the computer will then
perform the requested task based on the information it has gathered.
word In binary communications, multiple
bytes associated together are usually called a
word.
workgroup A group of individuals who
work together and share the same files and
databases over a local area network. Special
groupware such as Lotus Notes coordinates the
workgroup and allows users to edit drawings
or documents and update the database as a
group.
working directory Programs that need to
save temporary files or configuration data
while they are running do so within their
working directory. Users can also have a
working directory to save their temporary files.
workstation 1. In networking, any personal
computer (other than the file server) attached
to the network.
2. A high-performance computer optimized
for graphics applications such as computeraided design, computer-aided engineering, or
scientific applications.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com

Glossary

World Wide Web (WWW) This is the graphical extension of the Internet that features millions of pages of information accessed though
the use of the Hypertext Transfer Protocol
(HTTP).
write-back memory A type of memory
where, when changes are made to cached data,
the changes are not made to the original data at
the same time.
write-protect tab The small notch or tab in
a floppy disk that is used to write-protect it.
write-protect To prevent the addition or
deletion of files on a disk or tape. Floppy disks
have write-protect notches or small write-protect tabs that allow files to be read from the
disk, but prevent any modifications or deletions. Certain attributes can make individual
files write-protected so they can be read but not
altered or erased.
writing step The step in the EP print process
where the items being printed are written to the
EP drum. In this step, the laser is flashed on and
off as it scans across the surface of the drum.
The area where the laser shines on is discharged
to almost ground (100 volts).
write-through memory A type of cache in
which changes are made to the cached data are
simultaneously made to the original copy,
rather than being marked for later updating.
Also spelled write thru.
x86 series The general name given to the
Intel line of IBM-compatible CPUs.

655

XGA Acronym for Extended Graphics Array.

XGA is only available as a micro channel architecture expansion board; it is not available in
ISA or EISA form. XGA supports resolution of
1024 horizontal pixels by 768 vertical pixels
with 256 colors, as well as a VGA mode of 640
pixels by 480 pixels with 65,536 colors, and
like the 8514/A, XGA is interlaced. XGA is
optimized for use with graphical user interfaces, and instead of being very good at
drawing lines, it is a bit-block transfer device
designed to move blocks of bits like windows
or dialog boxes.
zero insertion force (ZIF) A type of processor socket where you dont have to snap
the chip into the socket. Rather, you simply set
the chip into the ZIF socket and push a bar
down to secure it.
zero wait state Describes a computer that
can process information without wait states. A
wait state is a clock cycle during which no
instructions are executed because the processor
is waiting for data from a device or from
memory.
ZIF socket Abbreviation for Zero Insertion
Force socket. A specially designed chip socket
that makes replacing a chip easier and safer.
zone In DNS, a zone is a section of
namespace. For example, ABCDEFG.com
would be a zone.
zone transfer Transfer of DNS information
from one server to another.

Xeon A member of the Pentium family, the

Xeon is a high-availability processor from Intel
that has been designed to work in a multiprocessing environment.

Copyright 2001 SYBEX, Inc., Alameda, CA

www.sybex.com