0% found this document useful (0 votes)

280 views

DC Fundamentals

gdgsdg

Uploaded by

johnny

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

280 views

DC Fundamentals

gdgsdg

Uploaded by

johnny

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 117

Fundamentals Series

Day One: Data Center

Fundamentals

Learn all the basics of data centers,

from components to cabling to
controllers, and how Juniper products
scale that technology. Its day one,
and you have a data center to build.

By Colin Wrightson

DAY ONE: Data Center FUNDAMENTALS

Day One: Data Center Fundamentals provides a thorough understanding of all the components that make up a data center solution using Juniper Networks products and
networking technologies. Youll learn how key data center principles fit together by
using an example architecture that is common throughout the book, providing an easy
reference point to gauge different data center solutions.
By books end, youll be able to design your own data center network and in the process come to understand why you would favor one technology or design principle over
another. The author points out subtle differences along the way, and provides links to
authoritative content that will help you with the technical specifications of the components, protocols, controllers, and configurations.
Data center architectures are evolving at an exponential rate. With the cloud reaching
maturity and SDNs rise from the trough of disillusionment towards the plateau of
profitability an update on the new fundamentals of data center architecture is long overdue. Colin Wrightson tackles the topic head-on in this excellent new addition to the Day
One library.
Perry Young, SVP, Tier-1 US Bank, Juniper Ambassador, JNCIP-SEC/SP/ENT, JNCDS-DC
This very timely book is essential reading if you want to keep up with the rapidly changing world of data centers. It takes you all the way from cabling to SDN technology, explaining all the fundamental principles in a well-written, easily digestible format.
Julian Lucek, Author and Distinguished Engineer, Juniper Networks
After years of remaining relatively static, the designs and technologies behind a data
center network are now evolving rapidly. The speed of change makes it difficult to keep
up with the latest architectures and technologies. Colin Wrightson, one of the data center wizards at Juniper Networks, has done of an amazing job of explaining these design
choices and technologies in a simple, easy-to-read book. Highly recommended for anyone who is considering implementing a new data center network.
Andy Ingram, VP Data Centers, Center of Excellence, Juniper Networks.
Juniper Networks Books are singularly focused on network productivity and efficiency.
Peruse the complete library at www.juniper.net/books.
ISBN 978-1-941441-39-8

51800

9 781941 441398

Day One: Data Center Fundamentals

By Colin Wrightson

Chapter 1: Common Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

Chapter 2 : Architectures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Chapter 3: Cabling. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Chapter 4: Oversubscription . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
Chapter 5: Fabric Architecture. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Chapter 6: IP Fabrics and BGP. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
Chapter 7: Overlay Networking. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
Chapter 8: Controllers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
Chapter 9: EVPN Protocol. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121

2016 by Juniper Networks, Inc. All rights reserved.

Juniper Networks and Junos are registered trademarks of
Juniper Networks, Inc. in the United States and other
countries. The Juniper Networks Logo, the Junos logo,
and JunosE are trademarks of Juniper Networks, Inc. All
other trademarks, service marks, registered trademarks,
or registered service marks are the property of their
respective owners. Juniper Networks assumes no
responsibility for any inaccuracies in this document.
Juniper Networks reserves the right to change, modify,
transfer, or otherwise revise this publication without
notice.
Published by Juniper Networks Books
Author & Illustrations: Colin Wrightson
Technical Reviewers: Anoop Kumar Sahu, Oliver
Jahreis, Guy Davies, Thorbjoern Zieger, Bhupen Mistry
Editor in Chief: Patrick Ames
Copyeditor and Proofer: Nancy Koerbel
ISBN: 978-1-941441-39-8 (paperback)
Printed in the USA by Vervante Corporation.
ISBN: 978-1-941441-40-4 (ebook)
Version History: v1, September 2016

About the Author

Colin Wrightson is a Consultant System Engineer with
the EMEA Center of Excellence team focusing on data
center product and design and has been with Juniper
Networks for over six years. His previous roles within
Juniper have been Systems Engineer and Senior Systems
engineer for enterprise covering government and defense
sectors. Prior to Juniper he worked for Cisco partners as
field engineering, engineering lead, and then pre-sales,
before seeing the error of his ways and joining Juniper.
Authors Acknowledgments
Id like to thank Patrick Ames who has spent far too long
correcting my questionable grammar and spelling with
the help of Nancy Koerbel. Thank you to the technical
reviewers. Id also like to thank Bhupen (social media
assassin) for his continued encouragement during this
process, Mark Petrou (big cod), who first gave me the
idea for a book, and Starbucks, whose coffee made a lot
of early morning writing easier. Last, but most importantly not least, I want to thank my long-suffering and
loving wife who makes all of this possible. Oh, and hello
to Jason Issacs.
http://www.juniper.net/dayone

Welcome to Day One

This book is part of a growing library of Day One books, produced and
published by Juniper Networks Books.
Day One books were conceived to help you get just the information that
you need on day one. The series covers Junos OS and Juniper Networks
networking essentials with straightforward explanations, step-by-step
instructions, and practical examples that are easy to follow.
The Day One library also includes a slightly larger and longer suite of
This Week books, whose concepts and test bed examples are more
similar to a weeklong seminar.
You can obtain either series, in multiple formats:
Download a free PDF edition at http://www.juniper.net/dayone.
Get the ebook edition for iPhones and iPads from the iTunes Store.

Search for Juniper Networks Books.

Get the ebook edition for any device that runs the Kindle app

(Android, Kindle, iPad, PC, or Mac) by opening your devices

Kindle app and going to the Kindle Store. Search for Juniper
Networks Books.
Purchase the paper edition at either Vervante Corporation (www.

vervante.com) or Amazon (amazon.com) for between $12-$28,

depending on page length.
Note that Nook, iPad, and various Android apps can also view

PDF files.
If your device or ebook app uses .epub files, but isnt an Apple

product, open iTunes and download the .epub file from the iTunes
Store. You can now drag and drop the file out of iTunes onto your
desktop and sync with your .epub device.

Audience and Scope of This Book

This book is intended for both Enterprise and service provider engineers,
network administrators, network designers/architects, and anyone who
wants to understand the basic principles of data center design. This
book provides field-tested solutions for common data center network
deployment scenarios, as well as the brief background information
needed to understand and design these solutions for your own environment.
The chapters of this book are organized in a logical sequence to help
provide the reader with a step-by-step understanding of data center
design principles and how these principles can then be developed into
a solution that fits the role of a modern data center.

What You Need to Know Before Reading This Book

Before reading this book, you should have a basic understanding of
network protocols and general design terms. While this book does not
cover Junos operating systems and configurations, there are several
excellent books on learning Junos in the Day One library at at http://
www.juniper.net/dayone.
This book makes a few assumptions about you, the reader:
You have a basic understanding of Internet Protocol (IP) versions

4 and 6
You have a basic understanding of network design at a campus

level
You work with servers and want to understand the network side

of data centers

What You Will Learn by Reading This Book

Basic principles of data center design and how they have evolved
How different data center designs affect applications
What overlay and underlay are, and why they are important
How Controller and Controllerless networks can improve Layer 2

scale and operations

A better understanding of Juniper data center products

vii

Preface
This Day One book provides you with a thorough understanding of all
the components that make up a Juniper Networks data center solution;
in essence, it offers a 10,000 foot view of how everything regarding
Junipers data center solution fits together. Such a view enables you to
see the value Juniper provides over other vendors in the same space by
glimpsing the big picture.
This books starts by covering the basics and slowly builds upon core
ideas in order to cover more complex elements. Design examples relate
back to an example architecture that is common throughout the book,
thus providing an easy reference point to gauge guideline solutions.
The idea is to allow you to design your own network and, in the
process, come to understand why you would favor one technology or
design principle over another. In order to do that, this book points out
subtle differences along the way:
Chapter One: Common Components starts with the various

common components (products) you can use and where they sit in
the design topology. This is important because of the differences
between merchant silicon and custom silicon, and because
different vendors have different approaches and those approaches
can affect the network and services you may want to implement.
Chapter Two: The Top-of-Rack or End/Middle-of-Row chapter

looks at the different architectures you can use, depending on

your server selection, contention ratio, and overall rack and
cabling design.
Chapter Three: Cabling is sometimes a neglected element of the

larger design, but it can have interesting repercussions if not

considered.
Chapter Four: Over Subscription is the first design point to think

about when designing a network. If you know the speeds at which

your servers are connecting, and the bandwidth they are going to
need, not just on day one but also ex-number of years from now,
then you can start the design process and select the right products
and speeds for your network.

viii

Chapter Five: This fabric architecture chapter looks at different

solutions for the connectivity of products, their management,

and how they move data across the data center.
Chapter Six: This chapter on IP Fabrics and BGP covers how

switches talk and transport traffic to each other. This connects to

IP Clos networks, where the protocol selection is a manual
implementation, whereas in Ethernet fabrics the vendor has done
this for you. This chapter also ties back to Chapter 5 on fabric
solutions.
Chapter Seven: Overlay networking focuses on the different

types of overlays supported, how they interact with the underlay

networking you may have just designed, and any considerations
you might take into account. This chapter looks at VTEPs, how
and where they terminate, and the different types you can
support.
Chapter Eight: This chapter on controller and controller-less

networks examines the benefits of using either a single pane of

glass or a static-based environment for both the network and the
servers, and explores whether or not there is a middle ground
using popular vendor offerings.
Chapter Nine: This chapter further explores a static-based

solution for overlay networks through the use of Ethernet VPN

(EVPN) to support Layer 2 overlay within the data center fabric.
Again, this book tries to simplify a complex networking entity so use
the links provided to update the precise specifications of the components, protocols, controllers, and configurations. A good place to start
is at the Juniper TechLibrary: http://www.juniper.net/documentation.
The author and editors worked hard to make this data center primer
nice and simple. Its a difficult thing to do, whether publishing or
network architecting. Three words nice and simple should be the
overriding basis of your design, no matter which Juniper technology
or service you use.
By the end of this book, you should be able to explain your data center
design within a few sentences. If you cant, then its probably too
complex of a design.
Colin Wrightson, October 2016

Chapter 1
Common Components

The first question to pose in any data center design is What switch
should I use and where? Juniper Networks, like other vendors,
produces data center switches that meet stringent specifications in
order to fit within particular segments of a network. This chapter
provides an overview of the Juniper Networks switching solution,
allowing you to understand and compare the different port densities,
form factors, and port capabilities of the devices. The book then
moves on explain the placement of devices in the network and the
different architectures they support.
NOTE

The architecture and different layers within a data center are described in more detail in subsequent chapters.

Switches Used In the Data Center

The QFX Series by Juniper Networks is specifically designed for the
data center. These switches address the need for low latency and high
availability, high port density, and the flexibility to support different
architectures. They are ideal data center switches, and at the time of
this writing, consist of the following product line: the QFX5100
Series, the QFX5200 Series, and the QFX10000 Series.
NOTE

MORE?

Download each products datasheet for up-to-date improvements

and modifications made after this book was published at http://
www.juniper.net/us/en/products-services/switching/qfx-series/.
For more detailed information about Junos Fusion and data centers,
see the Juniper Networks OReilly book, The QFX10000 Series, by
Doug Hanks: http://www.juniper.net/us/en/training/jnbooks/
oreilly-juniper-library/qfx10000-series/.

Day One: Data Center Fundamentals

The QFX5100 Series

The QFX5100 Series has five iterations, depending on the number and
type of ports each has, to meet various data center requirements.
QFX5100-48S

The QFX5100-48S is a 10-Gigabit Ethernet Enhanced Small FormFactor Pluggable (SFP+) top-of-rack switch with 48SFP+ ports and
sixQuad SFP+ (QSFP+) 40GbE ports. Each SFP+ port can operate as a
native 10Gigabit port or a 100MB/1Gigabit port when 1_Gigabit
optics are inserted. Each QSFP+ port can operate as either 40GbE
uplink ports or access ports. Each QSFP port can also operate as 4x
10GbE ports using a 4x 10 breakout cable. The QFX5100-48S provides
full duplex throughput of 1.44Tbps, has a 1U form factor, and comes
standard with redundant fans and redundant power supplies. The
switch is availiable with either back-to-front or front-to-back airflow
and with AC or DC power supplies. The QFX5100-48S can be used in
multiple architectures, such as:
A standalone switch
A spine or leaf in an IP Fabric (covered in later chapters)
A master, backup, or line card in a QFX Virtual Chassis (covered

later)
A spine or leaf device in a Virtual Chassis Fabric (VCF) (covered

later)
A satellite device in a Junos Fusion fabric (covered later)
QFX5100-48T

The QFX5100-48T is a tri-speed 100/1000/10Gb BASE-T top-of-rack

switch with 48 10GBASE-T access ports and six 40GbE QSFP+ ports.
Each QSFP+ port can operate as either an uplink port or an access port.
Each QSFP port can also operate as a 4x 10GbE port using a 4x 10
breakout cable. The QFX5100-48T provides full duplex throughput of
1.44Tbps, has a 1U form factor, and comes standard with redundant
fans and redundant power supplies. The QFX5100-48T can be used in
multiple architectures, such as:
A standalone switch
Leaf in a IP Fabric
A master, backup, or line card in a QFX Virtual Chassis
A leaf device in a VCF
A satellite device in a Junos Fusion fabric

Chapter 1: Common Components

QFX5100-24Q

The QFX5100-24Q is a 40-Gigabit Ethernet QSFP+ switch with 24

QSFP+ ports. Each QSFP+ port can operate as a native 40 Gbps port or
as four independent 10 Gbps ports. It has a 1RU form factor and comes
standard with redundant fans and redundant power supplies, and can
be ordered with either front-to-back or back-to-front airflow with AC
or DC power supplies. The QFX5100-24Q switch also has two module
bays for the optional expansion module, QFX-EM-4Q, which can add a
total of eight additional QSFP+ ports to the chassis, thus providing 32
ports of 40GbE or 104 logical ports when using 10G port breakout
cables. All ports on the QFX5100-24Q and QFX-EM-4Q can be
configured as either access ports or as uplinks, and the QFX5100-24Q
switch provides full duplex throughput of 2.56 Tbps. The QFX510024Q can be used in multiple architectures, such as:
A standalone switch
Spine or leaf in a IP Fabric
A master, backup, or line card in a QFX Virtual Chassis
A spine or leaf device in a VCF
A satellite device in a Junos Fusion fabric

QFX5100-96S

The QFX5100-96S is a 10-Gigabit Ethernet Enhanced Small FormFactor Pluggable (SFP+) top-of-rack switch with 96 SFP+ ports and
eight 40GbE Quad SFP+ (QSFP+) ports. Each SFP+ port can operate as
a native 10 Gbps port or as a 100MB/1 Gbps port. The eight QSFP+
ports can operate at native 40 Gbps speed or can be channelized into
four independent 10 Gbps port speeds taking the total number of
10GbE ports on the switch to 128. The QFX5100-96S switch has a 2U
form factor and comes as standard with redundant fans, redundant
power supplies, both AC or DC power support, and the option of either
front-to-back or back-to-front airflow. The QFX5100-96S can be used
in multiple architectures, such as:
A standalone switch
Spine or leaf in a IP Fabric
A member of a QFX Virtual Chassis
A spine or leaf device in a VCF
A satellite device in a Junos Fusion fabric

Day One: Data Center Fundamentals

QFX5100-24Q-AA

The QFX5100-24Q-AA is the same as the QFX5100-24Q with 24

QSFP+ ports of 40-Gigabit Ethernet. Each QSFP+ port can operate as
a native 40 Gbps port or as four independent 10 Gbps ports. It has a
1RU form factor and comes standard with redundant fans, redundant
power supplies, and can be ordered with either front-to-back or
back-to-front airflow with AC or DC power supplies. The QFX510024Q switch provides full duplex throughput of 2.56 Tbps. The
QFX5100-24Q-AA module bay can accommodate a single Packet
Flow Accelerator (PFA), a doublewide expansion module (QFX-PFA4Q), or two singlewide optional expansion modules as the standard
QFX5100-24Q.
The QFX-PFA-4Q, which features a high-performance field-programmable gate array (FPGA), provides four additional QSFP+ ports to the
chassis. This switch provides the hardware support to enable Precision
Time Protocol (PTP) boundary clocks by using the QFX-PFA-4Q
module. The QFX5100-24Q-AA also supports GPS or PPS in and out
signals when QFX-PFA-4Q is installed.
The CPU subsystem of this switch includes a two-port 10-Gigabit
Ethernet network interface card (NIC) to provide a high bandwidth
path or to alternate traffic path to guest VMs directly from the Packet
Forwarding Engine.
The QFX5100-24Q-AA can be used as a standalone switch that
supports high frequency statistics collection. Working with the Juniper
Networks Cloud Analytics Engine, this switch monitors and reports
the workload and application behavior across the physical and virtual
infrastructure.
The QFX5100-24Q-AA can be used as a top-of-rack switch where you
need application processing in the switch or Layer 4 services such as
NAT (Network Address Translation), packet encryption, load balancing, and many more services.
MORE?

The data sheets and other information for all of the QFX5100 Series
outlined above can be found at http://www.juniper.net/assets/us/en/
local/pdf/datasheets/1000480-en.pdf.

Chapter 1: Common Components

The QFX5200 Series

At the time of this books publication, the QFX5200 Series is comprised of a single device, the 32C.
QFX5200-32C

The QFX5200-32C is a 100Gigabit Ethernet top-of-rack switch that

supports 10/25/40/50 and 100GbE connectivity, allowing this 1RU
switch to support the following port configurations:
32 ports of 100GbE
32 ports of 40GbE
64 ports of 50GbE (using a breakout cable)
128 ports of 25GbE (using a breakout cable)
128 ports of 10GbE (using a breakout cable)

The QFX5200-32C comes standard with redundant fans and redundant power supplies supporting either AC or DC, and is available with
either frontto-back or back-to-front airflow. The QFX5200 can be
used in multiple architectures, such as:
A standalone switch
Spine or leaf in a IP Fabric
A satellite device in a Junos Fusion fabric
MORE?

The datasheets and other information for all of the QFX5200 Series
outlined above can be found at http://www.juniper.net/us/en/productsservices/switching/qfx-series/

Day One: Data Center Fundamentals

The QFX10000 Series

The QFX10000 Series is comprised of four platforms at the time of
this books publication: the fixed format QFX10002-36Q and
QFX10002-72Q, and the chassis-based QFX10008 and QFX10016.
QFX10002-36Q

The QFX10002-36Q is a 100Gigabit Ethernet aggregation and spine

layer switch that supports 10/40 and 100GbE connectivity, allowing
this 2RU switch to support the following port configurations:
12 ports of 100GbE
36 ports of 40GbE
144 ports of 10GbE (using a breakout cable)

The QFX10002-36Q comes standard with redundant fans and

redundant power supplies supporting either AC or DC, and can be
ordered with front-to-back airflow. The QFX10002-36Q series and
can be used in multiple architectures, such as:
A standalone switch
Leaf or spine layer switch in a IP Fabric
A aggregation device in a Junos Fusion fabric
QFX10002-72Q

The QFX10002-72Q is a 100Gigabit Ethernet aggregation and spine

layer switch that supports 10/40 and 100GbE connectivity. Allowing
this 2RU switch supports the following port configurations:
24 ports of 100GbE
72 ports of 40GbE
288 ports of 10GbE (using a breakout cable)

The QFX10002-72Q comes standard with redundant fans and

redundant power supplies supporting either AC or DC, and can be
ordered with front-to-back airflow. The QFX10002-72Q series can be
used in multiple architectures, such as:
A standalone switch
Leaf or spine layer switch in a IP Fabric
A aggregation device in a Junos Fusion fabric

Chapter 1: Common Components

MORE?

Complete, up-to-date details on the two fixed QFX10000 Series

switches can found here: http://www.juniper.net/assets/us/en/local/pdf/
datasheets/1000529-en.pdf.
QFX10008

The QFX10008 is a chassis-based high-density aggregation and spine

layer switch that supports 1/10/40 and 100GbE connectivity across
eight slots, allowing this 13RU switch to support the following port
configurations:
240 ports of 100GbE
288 ports of 40GbE
1152 ports of 10GbE

The QFX10008 supports three line card options:

The QFX10000-36Q line card provides 36 ports of 40GbE,12

ports of 100GbE, and 144 ports of 10GbE with breakout cables

The QFX10000-30C line card provides 30 ports of 40GbE and

30 ports of 100GbE
The QFX10000-60S-6Q line card provides 60 ports of 1/10GbE

with six ports of 40GbE and 2 ports of 100GbE

The QFX10008 comes standard with redundant route engines (supervisor cards in Cisco speak) in a mid-planeless architecture that supports dual redundant fan trays, with six redundant power supplies
supporting either AC or DC, and can be ordered with front-to-back or
back-to-front airflow. The QFX10008 series can be used in multiple
architectures, such as:
A standalone switch
Leaf, spine, or aggregation layer switch in a IP Fabric (covered in

later sections)
A aggregation device in a Junos Fusion fabric (covered in later

sections)
QFX10016

The QFX10016 is a chassis-based aggregation and spine layer switch

that supports 10/40 and 100GbE connectivity across 16 slots. This
21RU switch supports the following port configurations:
480 ports of 100GbE
576 ports of 40GbE
2304 ports of 10GbE

Day One: Data Center Fundamentals

The QFX10016 supports three line card options:

The QFX10000-36Q line card provides 36 ports of 40GbE, 12

ports of 100GbE, and 144 ports of 10GbE

The QFX10000-30C line card provides 30 ports of 40GbE and

30 ports of 100GbE
The QFX10000-60S-6Q line card provides 84 ports of 10GbE,

six ports of 40GbE, and two ports of 100GbE

The QFX10016 comes standard with redundant route engines in a
mid-planeless architecture that supoprts dual redundant fan trays,
with ten redundant power supplies supporting either AC or DC, and
can be ordered with front-to-back airflow. The QFX10016 series can
be used in multiple architectures, such as:
A standalone switch
Leaf, spine, or aggregation layer switch in a IP Fabric (covered in

later sections)
A aggregation device in a Junos Fusion fabric (covered in later

sections)
MORE?

Up-to-date details on the chassis-based QFX10000 series switches can

found here: http://www.juniper.net/assets/us/en/local/pdf/
datasheets/1000529-en.pdf.

MORE?

For more detailed information about the QFX Series and data centers,
see the Juniper/OReilly book, The QFX10000 Series, at http://www.
juniper.net/books.
With so many iterations of the QFX Series, its worth discussing the
different silicon types used in these products as it can have a bearing on
their placement in the data center and their capabilities once installed.

Chapter 1: Common Components

Custom and Merchant Silicon

When silicon is discussed in this book, it means the Packet Forwarding Engine or ASIC (application-specific integrated circuit)
that a switch uses to process L2 packets. Merchant silicon describes
the use of off the shelf third-party Packet Forwarding Engines
that can be implemented into a networking product, whereas
custom silicon relates to a company, like Juniper, that designs their
own Packet Forwarding Engines and implements them into their
own products.
Merchant silicon is used throughout the industry by multiple
switch vendors and for good reason its an efficient method of
bringing new switches and capabilities to market in a shorter
timeframe than a custom silicon product. But there are always
trade-offs to consider with a mass-market product.
For example, the QFX5100 and 5200 Series devices both use
Broadcom Packet Forwarding Engines. Specifically, the QFX5100
uses the Broadcom T2 chipset and the QFX5200 uses the newer
Broadcom Tomahawk Chipset. This is no different than Cisco,
Arista, HP, Dell Force 10, Huawei, and others who use Broadcom
in their versions of similar switches.
These are excellent switches with high throughput, high port
densities, and with a software feature set that provides 90% of
what most networks need to do with a switching device. But the
trade-off is that these are generic devices that have a low buffer
capability, a default set of features that may not be able to be
enhanced outside of the parameters that are supported in that
hardware, and, given how fast these products come to market, they
could be out of date very quickly.
The other option is to use a product based on custom silicon, with a
chipset such as the one found in the QFX10000 Series. With
custom-made silicon, youll get a higher level of innovation that
will allow scaling and newer services to be introduced in software
rather than hardware. This in turn will allow you to be more
flexible in supporting a greater number of different port speeds and
other hardware-related features, higher buffer capabilities, and a
more flexible software feature set allowing it to be placed at
different areas of the network.
But the trade- off is that custom silicon products will be more
expensive due to the higher level of investment and research that
has gone into their development. This means the product will be on

Day One: Data Center Fundamentals

the market longer than a merchant silicon version (to recoup the initial
production costs) and that you need to consider future technology
shifts that may happen and their effects on both types of products.
There are pros and cons to both approaches, so I suggest you consider
using both merchant and custom silicon, but in different positions
within the network to get the best results.
Network designers tend to use the following rule of thumb: use
merchant silicon at the leaf/server layer where the Layer 2 and Layer 3
throughput and latency is the main requirement, with minimal buffers,
higher port densities, support for open standards, and innovation in
the switches OS software. Then, at the spine or core, where all the
traffic is aggregated, custom silicon should be used, as the benefits are
greater bandwidth, port density, and larger buffers. You can also
implement more intelligence at the spine or core to allow for other
protocols such as EVPN for Data Center Interconnect (DCI), analytics
engines, and other NFV-based products that may need more resources
than are provided on a merchant silicon-based switch.
IMPORTANT

The most important aspect in this requirement is making sure the

products you have selected have all the required features and performance capabilities for the applications you need. Always seek out the
most current specifications, datasheets, and up-to-date improvements
on the vendors web site. For up-to-date information about Juniper
switching platforms, you can go to: http://www.juniper.net/us/en/
products-services/switching/.

Chapter 2
Architectures

Now that you have a better understanding of the Juniper products

for the data center, lets move on to the different types of physical
architectures that can be deployed. The physical architecture is
defined as the physical placement of your switches in relation to
the physical rack deployment in the data center you are either
considering or have already implemented.
There are two main deployment designs, either top-of-rack or
end-of-row.

Top-of-Rack
In top-of-rack (ToR) designs one or two Ethernet switches are
installed inside each rack to provide local server connectivity.
While the name top-of-rack would imply placement of the switch
at the top of the physical rack, in reality the switch placement can
be at the bottom or middle of the rack (top-of-rack typically
provides an easier point of access and cable management to the
switch).

Day One: Data Center Fundamentals

Figure 2.1

ToR Switch Positions

The switches deployed at the top-of-rack are typically of a fixed

format, 1RU to 2RU in height. The QFX5100, QFX5110 and
QFX5200 switches are examples of switches developed for this
requirement.
The main benefit of positioning your switches within racks is the ease
of server-to-switch connectivity, removing the need for long cable runs,
while allowing 1GbE copper, and 10 or 40GbE fiber/DAC cables to be
run directly from the servers to the switch. This style of implementation eliminates the need for cabling patch panels, which requires
additional racks, and running large quantities of cabling between
racks. Top-of-rack eliminates the issues around long cable runs and
makes replacing cables easier when faster network speeds are required.
With a top-of-rack design, each rack can be managed as a single entity
within the data center. This can make the process of server swaps and
upgrades easier because they can be implemented on a rack-by-rack
basis without affecting other servers and related applications in other
racks. It also means that upgrades on the switches can be done in a
managed and staged approach, whereby once the top switches are
upgraded first and traffic is again flowing, the secondary switch in the
rack can be upgraded and so on. The implementation of in-service
software upgrades (ISSU) in top-of-rack switches removes this need
altogether, but as youll see in Chapter 5 on fabric architecture, there
are usually some caveats.

Chapter 2: Architectures

Connectivity from the top-of-rack switch to the aggregation layer is

usually via dual 10 or 40GbE fiber connections for resiliency, as shown
below in Figure 2.2.

Figure 2.2

ToR to Aggregation Uplinks

For uplink cables, fiber provides a considerably better investment for

long-term deployments than copper cabling, as it can carry higher
bandwidths over a longer distance and provide the flexibility to
support higher bandwidth when higher speeds are required.
With this in mind the top-of-rack switch you select should be able to
support flexible uplink port speeds. An example would be the
QFX5100-48S, which has six 40GbE uplink ports, but each of those
uplink ports can be broken down into four individual 10GbE links. So
while your requirements on day one may only be for 10GbE uplinks,
you know you have the ability to support 40GbE in the future without
the need to swap out expensive hardware.
Arguably, one of the disadvantages of the top-of-rack design is the
expansion in size of your management domain, as every switch is an
individually managed device with all of the configuration and software
burdens that come with that device. For example, if you have a data
center with ten rows of ten racks, with two switches per rack, thats
200 switches that need to be managed and monitored. While a lot of

Day One: Data Center Fundamentals

the configuration can be duplicated for the majority of switches, that

still represents a lot of overhead from a management point of view and
exposes many potential points in the network to misconfiguration.
NOTE

Juniper has addressed these concerns by providing virtualization

technologies like Virtual Chassis Fabric and Junos Fusion to simplify the
management domain. This simplification is achieved by implementing a
virtualized control plane over a large number of switches, effectively
creating a large virtual switch where each physical switch is a virtual
line card. You then have the flexibility to implement your virtual switch
over several racks, or rows of racks, and provide faster implementation
with a single point of configuration over virtual pods or clusters of
network switches. These technologies are discussed in more detail in
Chapter 5 of this book: Fabric Architecture.
Another disadvantage in top-of-rack design is in the number of ports
you might waste. The average top-of-rack switch comes with 48 ports
of 1/10GbE. With two switches per rack, that provides 96 ports per
rack. You would need a lot of servers per rack to utilize all of those
connections. This is where an end-of-row solution provides an advantage as you waste fewer ports but increase the number and size of
inter-rack cable connections.
There are ways around the port wastage, such as cross connecting
servers between racks. While not a great idea from a cable management
point of view, cross connecting does allow you to better utilize your
switch ports and provide resiliency for servers. As the diagram in Figure
2.3 illustrates, a 48-port switch would be positioned in each rack, 24
ports are dedicated to the servers in that rack and the other 24 are
dedicated for the rack next to it.

Figure 2.3

ToR Cross Connect

Chapter 2: Architectures

In the other rack the same is done again and 50% of the server connections cross into the other rack and vise versa. As mentioned earlier, this
is not the most elegant of designs, and you would need to factor in
extra cable length to bridge the gap between racks, but it is more
practical and less expensive than a two switch per rack design.
And this situation brings up a good point: not every data center
installation or design is perfect, so if the last example works for you
and you are aware of its limitations and potential issues, then implement accordingly.
Table 2.1

Summary of ToR Design

Advantages

Limitations

Cabling complexity is reduced as all the servers are

connected to the switches located in the same rack
and only fiber uplink connections pass outside the
rack.

There may be more unused ports in each rack and it

is very difficult to accurately provide the required
number of ports. This results in higher un-utilized
ports/ power/ cooling.

If the racks are small, there could be one network

switch for two-three racks.

Unplanned Expansions (within a rack) might be

difficult to achieve using the ToR approach.

ToR architecture supports modular deployment of

data center racks as each rack can come in-built
with all the necessary cabling/ switches and can be
deployed quickly on-site.

Each switch needs to be managed independently. So

your CapEx and OpEx costs might be higher in ToR
deployments.

This design provides scalability, because you may

require 1GE/ 10GE today, but you can upgrade to
10GE/ 40GE in the future with minimum costs and
changes to cabling.

1U/2U switches are used in each rack, achieving

scalability beyond a certain number of ports would
become difficult.

End-of-Row
The end-of-row design (EoR) was devised to provide two central
points of aggregation for server connectivity in a row of cabinets as
opposed to aggregation within each rack as shown in the top-of-rack
design. Each server within each cabinet would be connected to each
end-of-row switch cabinet either directly via RJ45, via fiber, or if the
length is not too great, with DAC or via a patch panel present in each
rack.

Day One: Data Center Fundamentals

Figure 2.4

EoR Design

For a redundant design there might be two bundles of cables to each

rack, each running to opposite end-of-row network racks. Within the
server rack the bundle of cable is normally wired to one or more patch
panels fixed to the top of the cabinet
The individual servers use a relatively short RJ45/DAC/Fiber patch
cable to connect from the server to the patch panel in the same rack.
The bundles of cable from each rack can then be channeled through
either overhead cable troughs or under floor bundles to the end-of-row
network racks. Depending on how much cabling is required, its
common to have a rack dedicated to patching all of the adjacent cables,
or within the rack that contains the end-of-row network switch.
It is also quite common to directly connect the server cables to the
end-of-row switches as opposed to via a patch panel, but again this is
dependent on your cabling policy, distance, and cable bundle sizes.

Chapter 2: Architectures

Another version of this design is referred to as middle-of-row which

involves bundles of cables from each server rack going to a pair of
racks positioned in the middle of the row. Both designs are valid but
careful consideration needs to be taken concerning the cabling design
for either design can end up in a mess.
The switch positioned in either the end-of-row or middle-of-row is
generally a chassis-based model. A chassis-based switch would provide
a higher density of network connections and possibly a higher level of
availability, as the chassis would have multiple power supplies, dual
routing engines, and multiple fabric cards. This is the case with most
chassis-based solutions from Juniper if you need faster connection
rates then all you need to do is upgrade the physical line cards in the
device, not the whole chassis, allowing core components to last a lot
longer than a typical top-of-rack switch at a lower OPEX cost over the
lifetime of the device.
From a management point of view, your management domain is
considerably reduced because it is based on an entire row rather than
the top-of-rack design where your domain is per rack. However, it does
mean that your failure domain increases to encompass all of the servers
in that row, and that upgrades need to be better planned because the
effect of a failure would have a larger impact than you have in the
top-of-rack design (and it removes the top-of-rack option to upgrade
certain racks based on common applications or services).
One benefit, which is discussed in Chapter 4, is that your over-subscription ratios or contention ratios will be potentially better in an
end-of-row design as compared to a top-of-rack design. For example,
what if you were terminating 48 ports of 10GbE per line card and you
wanted to keep a 3:1 ratio of server to uplink traffic? Several years ago
this would involve terminating all of your 10GbE connections over
several line cards and then pushing this traffic to the backplane and a
dedicated uplink line card. You would have had to provision 160GbE
of uplink capacity, or 4x 40GbE, to hit that ratio and then it would
only have been for those 48 ports. If you need to terminate more than
that number of 10GbE connections you would need more than 40GbE.
Today, with mixed speed line cards, you can use a 60x 10GbE line card
that also has an additional 6x 40GbE or 2x 100GbE, allowing you to
hit a lower contention ratio on the same card without sending traffic
via the backplane, and thats traffic dedicated to that line card, which
means you can replicate it over every line card.

Day One: Data Center Fundamentals

Table 2.2

Summary of EoR Design

Advantages

Limitations

Since Chassis Switches are deployed in EoR

configurations, expansion (for the total number
of ports) can be done by just adding a line card as
most of the Chassis Switches are planned for
expandable configurations, without limiting
backplane capacities.

The EoR architecture would bring down the

number of un-utilized switch ports, drastically.
This decreases the capital expenditure, running
costs, and time needed for maintenance.

Chassis Switches enable a high availability

configuration with almost no single points of
failure, as most of the critical components (control
module, cooling module, power supply module,
etc.) can be deployed in redundant (1+1 or N+1 or
N+N) configurations. The failover is almost
immediate (often without affecting end user
sessions).

Longer cables are required to connect the Chassis

switch (at end of the row) to each server, in EoR
deployments and hence special arrangements might
be required to carry them over to the aggregation
switch. This might result in excessive space
utilization at the rack/ data center cabling routes,
increasing the amount of data center space required
to store the same number of servers.

Placement of servers can be decided independently,

without any minimum/maximum servers in a
single rack constraints. So, servers can be placed
more evenly across the racks and hence there may
not be excessive cooling requirements due to
densely placed servers.

Each switch needs to be managed independently.

So your CapEx and OpEx costs might be higher in
ToR deployments.

Since each packet fewer switches to pass through,

the latency and delay involved in passing through
multiple switches is minimized.

The cost of higher performance cables (used in data

centers) can be considerable and hence cabling costs
can get higher than ToR deployments. Also, fault
finding on cabling issues (especially if cabling needs
to be replaced) becomes more intrusive and adds to
overall OpEx.

The EoR architecture would drastically reduce the

number of switch ports that are not being utilized.
This decreases the CapEx, running costs, and time
needed for maintenance.

Its difficult, and more expensive, to upgrade cabling

infrastructure to support higher speeds or
performance, as lengthier cables need to be replaced
individually when upgrading from 10GE to 40GE,
for example.

Servers that exchange a considerable amount of

data packets between each other can be connected
to the same line card in the Chassis switch,
regardless of the rack they belong to. This
minimizes delay and enables better performance
due to local switching.

Chapter 3
Cabling

The last chapter on top-of-rack and end-of-row placement of the

components brought to light the importance of data center physical
cabling design. When designing a cabling infrastructure, the physical
layout of the cabling plant, the signal attenuation and distance, as
well as installation and termination, require thorough consideration.
Investing in the optimal cabling media to support not only your day
one requirements, but also the need for higher media speeds such as
40, and 100 Gigabit Ethernet (GbE) connectivity involves striking a
balance among bandwidth, flexibility, and cost designed for your
purposes.
Most component vendors provide three supported options: copper,
fiber, and direct attach copper (DAC). Lets discuss each and how
you might implement any of them into your design.

Day One: Data Center Fundamentals

Copper
Copper cabling is often referred to as Cat5e and Cat6, which refers to
RJ45 terminated cabling over a twisted pair that provides support for
10/100/1000Mb and 10GbE connection speeds over a certain frequency up to 100m. See Table 3.1 below for detailed information.
In the context of a top-of-rack design solution, one would expect
copper to be implemented within each rack. In an end-of-row design
solution the copper cabling could be implemented between the servers
and switches, but the distance would need to be considered as well as
the size of the cabling plant. Also, any future upgrades might mean
removing this cabling to support higher speeds, which is always a
consideration.
But if this is whats required, then the QFX5100 series supports it in
the form of the QFX5100-48T switch. You also have the option of
using the EX4300 and EX3400 series.
Table 3.1 provides an overview of the different connector types, cable
standards, cabling, distances each cable can cover, the frequency
typically used, and where in the design the cable is used, as well as the
Juniper Networks supported switch series, the EX or QFX Series.
Table 3.1

Copper Cabling Options

Connector/
Media

IEEE Cable Standard

Max Distance

Frequency
(MHz)

Recommended
Placement and
Supported Switch

1000BASE-T
Copper

CAT5e
CAT6

328ft (100m)

1-100
1-250
1-500

ToR in rack

1-100
1-250
1-500

ToR in rack

CAT6A (STP/UTP)
100BASE-TX

CAT5e
CAT6

328ft (100m)

CAT6A (STP/UTP)
10GBASE-T

CAT7

328ft (100m)

1600

CAT6A

328ft (100m)

1500

CAT6 (UTP)

98ft (30m)

1250

CAT 6 (STP)

98ft (30m)

1250

QFX5100-48T
EX4300 and EX3400

QFX5100-48T

Chapter 3: Cabling

Fiber
There are two classifications for optical fiber: single-mode fiber-optic
(SMF) and multimode fiber-optic (MMF).
In SMF, light follows a single path through the fiber while in MMF it
takes multiple paths resulting in a differential mode delay or (DMD).
In the context of a top-of-rack design solution one would expect MMF
to be used within each rack and between connecting racks up to the
aggregation layer, if the distance is relatively short (up to 400m). In an
end-of-row design solution it could be implemented between the
servers and switches. If distance permits, then MMF could also be
implemented up to a core or aggregation layer, and if not, then then
you could implement SMF.
All of the switches in the QFX Series support fiber interfaces. To make
things a little easier Table 3.2 lists an overview of the different connector types, cable standards, cabling types, the distances each can cover,
the wavelengths used, where in the design the cable is used, and the
Juniper Networks supported switch series.

Table 3.2

Fiber Cabling Options

Connector/
Media

IEEE Cable
Standard

Fiber Grade & Max

Distance

1000BASE-T
RJ45 Copper SFP

RJ-45

328ft (100m)

1000BASE-SX

LC-MMF

1000BASE-LX

10G-USR

Wavelength

Recommended
Placement and
Supported Switch

ToR in rack
QFX5100-48S & 96T

LC SMF

LC- MMF

FDDI - 220m (721 ft)

1-160

ToR & EoR

OM1 - 275m (902 ft)

1-200

QFX5100 & EX Series

500m (1640 ft)

1-400

OM2 - 550m (1804 ft)

1-500
ToR & EoR within rack

SMF 10Km (6.2m)

OM1 550m (1804 ft)

1-500

500m (1640 ft)

1-400

OM2 - 550m (1804 ft)

1-500

OM1 10m (32 ft)

840nm to
860nm

OM2 30m (98 ft)

OM3 100m (328 ft)

QFX5100 Series

ToR & EoR in & out of

rack
QFX5100 &
QFX10000 Series

Day One: Data Center Fundamentals

10GBASE-SR

LC-MMF

FDDI 26m (85 ft)

OM1 - 33m (108 ft)

840nm to
860nm

ToR & EoR for both in

rack and out of rack.
QFX5100 &
QFX10000 Series

66m (216 ft)

OM2 - 82m (269 ft)
OM3 - 300m (984 ft)
OM4 400m (1213 ft)
10GBASE-LR

LC SMF

10km (6.2 miles)

ToR & EoR out of rack

1530nm to
1565nm

Long Distance DCI

QFX5100 &
QFX10000 Series

10GBASE-ER

LC-SMF

40GBASE-SR4

OM3 - 100m (325 ft)

12-ribbon
multimode fiber OM4 150m (425 ft)
crossover cable
MMF

840nm to
860nm

ToR & EoR out of rack

OM3 - 300m (984 ft)

12-ribbon
multimode fiber OM4 400m (1312 ft)
crossover cable
- MMF

840nm to
860nm

ToR & EoR out of rack

LC SMF

See note *

ToR & EoR out of rack

40GX10G-ESR4

40GBASE-LR4

40km (24.85 miles)

1260nm to
1355nm

10 Km (6.2 miles)

QFX5100 &
QFX10000 Series

40G-LX4

Dual LC SMF
& MMF

OS1 2 KM

LC SMF

2 Km (1.24 miles)

See note *

OM3 100m (328 ft)

QFX5100 &
QFX10000 Series

OM4 150m (492 ft)

40G-IR4

ToR & EoR out of rack

IEEE Cable
Standard

Max Distance

Minimum Cable
Bend Radius

Recommended
Placement and
Supported Switch

QFX-SFP-DAC-1M
10GbE

SFP permanently
attached.

1 m (3.3 ft)

1 in. (2.54 cm)

ToR in rack
QFX5100 &
QFX10000 Series

QFX-SFP-DAC-2M
10GbE

SFP permanently
attached.

2 m (6.6 ft)

1 in. (2.54 cm)

ToR in rack
QFX5100 &
QFX10000 Series

QFX-SFP-DAC-3M
10GbE

SFP permanently
attached.

3 m (9.9 ft)

1 in. (2.54 cm)

ToR in rack
QFX5100 &
QFX10000 Series

ToR in rack
QFX5200 &
QFX10000 Series

QFX5100 &
QFX10000 Series

There are also a few other choices of cable types at 40GbE, including
Active Optical Cables, which can go up to 30 meters and can provide
for another option in both top-of-rack and end-of-row solutions.
Another solution, if you dont require 40GbE on day one, is to use
40GbE to 4x 10GbE breakout cables. These allow you to break a
native 40GbE interface into 4x 10GbE SFP or DAC interfaces.
If there were a best practice design, then it would be as simple as DAC/
Active Optical Cables and RJ45 within racks, and fiber between racks.
MORE? Details on the cables supported for all of the QFX products mentioned

in this chapter can be found here: http://pathfinder.juniper.net/hct/

category/catKey=100001&pf=QFX+Series&standard=Ethernet.

Chapter 4
Oversubscription

This Day One book defines oversubscription as the maximum

throughput of all active southbound connections divided by the
maximum throughput of the northbound connections, or, in plain
language, if you have 20 servers each running at 10GbE connecting to
a single switch, thats 200GbE in combined (southbound) connectivity.
If you have 2x 40GbE uplink connections to the next layer in the
network on the same switch, thats 80GbE of northbound connectivity
(200 divided by 80 equals 2.5). So 2.5 is the oversubscription ratio.
Some people (generally server teams) may see this as a failure of the
network (not being able to provision 1:1), but servers rarely run at
100% bandwidth, so an element of oversubscription is fine and is
often required, unless money is no object. Working with your server
and application teams to define an acceptable ratio and design the
network accordingly is critical to your data center design.
And thats the purpose of this chapter, to take what you have learned
in the previous chapters and start applying it to this books default
design for instructional purposes.
Before moving on to that design phase, there are two more points to
touch on: switch oversubscription and network oversubscription.
Switch oversubscription occurs when the overall internal switching
bandwidth of the switch is less than the total bandwidth of all ingress
switch ports. So, if you have a 48-port switch with every port supporting 10GbE, you would have 480GbE in switch capacity. If the internal
bandwidths on the switch can only switch 240GbE of traffic at any
one time, then you have a 2:1 oversubscription ratio.
While top-of-rack 1RU switches from most vendors are now line rate
(Juniper switches have always been line rate), you do see oversubscrip-

Day One: Data Center Fundamentals

tion from the line card to the backplane on chassis-based switches in an

end-of-row design. Always check the backplane switch capacity in both
directions, as some vendors like to state a figure that you have to divide
by two in order to get the real bandwidth.
And as described in the first paragraph of this chapter, network oversubscription refers to a point of bandwidth consolidation where the
ingress or incoming bandwidth is greater than the egress or outgoing
bandwidth, thus getting the oversubscription ratio to a degree that
works for you and the applications you need to support.

Oversubscription Design
The starting point in designing any network is to understand the
requirements that have been set and design from that point forward.
To make this hypothetical as real as possible and provide a foundation
for the rest of the book, lets consider the following client requirements:
The client has two greenfield data centers (DC1 & DC2) that are 40Km
apart and will be connected to each other via the clients MPLS WAN
currently running on Juniper MX Series.
DC1:

DC1 will have five rows of 10 racks

Each rack in DC1 will house 14 servers
Each server has 3x 10GbE Fiber connections plus a single 1GbE

RJ45 management connection

Client would like a spine and leaf-based top-of-rack solution
Oversubscription ratio of 4:1 or lower (lower is preferred)
DC2:

DC2 will also have five rows of 10 racks

Each rack in DC2 will house five blade chassis with 4x 10GbE per

chassis
Client would like an end-of-row/middle-of-row solution
Oversubscription ratio of 1:4 or lower (lower is preferred)

Both data centers should be provisioned with support for EVPN to

allow Layer 2 stretch and possible support for multi-tenancy in the
future.
With these requirements outlined, lets begin to work out our subscription ratio and product placement.

Chapter 4: Oversubscription

DC1 Design
Starting with DC1, each rack houses 14 servers. Each server has 3x
10GbE fiber connections plus 1x 1GbE RJ45 connection for out of
band (OoB).
Total per rack = 3 x 14 (number of 10GbE times the number of

servers) = 42, plus 14 x 1GbE = 14

Total per a rack is 42 x 10GbE plus 14 x 1GbE
Per a row 42 x 10 (number of 10GbE per rack times the

number of racks per a row) = 420

Plus 14 x 10 (number of 1GbE times the number of racks per a

row) = 140
NOTE

The ratio of 10GbE and 1GbE per a row was used just in case end-ofrow is offered instead of top-of-rack at DC1.
From a subscription point of view, a rack has a total bandwidth of 420
GbE. With a 4:1 subscription ratio that would be 420 / 4 (total bandwidth divided by the subscription ratio) = 105 GbE as the uplink
capacity.
To hit the 105GbE mark or better you need to use 40GbE uplinks. You
could go with 3 x 40GbE, which would equal 120GbE, which would
lower the subscription ratio to 5:3, but this would mean three spine or
aggregation layer switches to terminate the uplinks, which is never
ideal. If you propose 4 x 40GbE per a top-of-rack switch, then you
have 160GbE of uplink capacity, which would give us a subscription
ratio of 2.65:1. This would also mean you could either have two, or
four, spine-layer switches per row and either 2 x 40GbE per spine layer,
or if we go with four spine-layer switches, then a single 40GbE connection to each.
Once you know what your interface specifications are you can match
them against the available products. The QFX5100-48S provides 48
ports of 10GbE and 6 x 40GbE uplinks, so its perfect.

NOTE

At this point a good designer should ask the client about resiliency.
Would a single switch per rack be okay, or would the client rather have
two switches per rack, which would increase cost and waste ports?
This is where an end-of-row solution could be a more cost-efficient
answer, as you get both the resiliency and better port utilization. For
the sake of brevity, lets assume that the client is happy with the single
top-of-rack per rack and a top-of-rack-based architecture.

Day One: Data Center Fundamentals

The rack design will look similar to Figure 4.1:

Figure 4.1

This Books Rack Design

Figure 4.1 includes a switch at the bottom of the rack to terminate the
OoB 1GbE connections from the servers. Best practice would suggest
that the OoB connection on the back of the top-of-rack switch is also
connected to this bottom switch, thus providing access to the switch
outside of the normal data path while allowing network management
traffic and analytics to flow without interruption.
Lets now move on to the spine layer, or aggregation points, per row.
The number of 40GbE connections per a top-of-rack switch is four, and
there are ten racks, so our total number of 40GbE connections per a
row is 40 x 40GbE. You have the option at this point to either position
two spine layer switches, so 20 x 40GbE per spine, or four, which would
take that down to 10 x 40GbE per spine. But in order to make this
decision you have to work out the subscription ratio of each row.
There are two ways of approaching doing this, either oversubscribe by a
factor or two again, or, try and keep the same subscription ratio as
prescribed for the leaf to spine.
If you reduce the 400GbE by a factor of two you would have to provision 200GbE of uplink capacity from the spine to the core. If you
propose two spines, that would be 100GbE per a spine switch, and if
you propose four spines, that would be 50GbE per spine.

Chapter 4: Oversubscription

With two spines you could propose 3 x 40GbE per spine or with four
spines you could propose 2 x 40GbE. In each of these cases you would
still be below the initial subscription ratio.
The other option is to propose 100GbE uplinks to the core. With a
two-spine solution you could propose 100GbE per spine and with a
four-spine design you could propose a 1 x 100GbE per spine, and, in
doing so, you would you could keep the same subscription ratio as
defined between the leaf and the spine. So the client would see no drop
in bandwidth northbound of the top-of-rack switches within each rack.
From a QFX Series point of view, if you go down the 40GbE route then
todays choice is the QFX5100-24Q, with its 24 ports of 40GbE plus
additional expansion. But if you want the option to do both 40GbE and
100GbE, then the QFX10002, with its flexibility to support both
solutions, and its built-in upgrade path from 40GbE to 100GbE, would
be the better option. The QFX10002 would also provide additional
benefits to the client in regard to its buffer capacity of 300Mb per a port
and analytics feature set.
The other option would be the QFX5200 model that also has 40 and
100GbE connectivity, but has a much smaller buffer capacity than the
10K. In each case it provides you with options to present back to the
client.
The choice to present either 200GbE or 400GbE uplink capacity from
each row to the client, based on their specifications, which would mean
a four-rack design per row, would look like Figure 4.2.

Figure 4.2

DC1 Proposed Data Center Rack Design Spine and Leaf

Day One: Data Center Fundamentals

And the logical subscription design of Figure 4.2 would look like the
one in Figure 4.3, below.

Figure 4.3

Oversubscription Design of DC1 Spine and Leaf

Now that the number of uplinks from each row is known, you can
work out what your core layer product should be. The core layer acts
as the point of aggregation between the rows of spine and leaves while
providing onward connectivity to the WAN.
As outlined previously, there are five rows of ten racks. Each row of
racks will have either 200GbE or 400GbE of uplink connectivity. So, at
a minimum, you would have to connect 10 x 100GbE links per core.
But, it would also be prudent to make sure you can support the higher
capacity, if at some stage the client prefers or wants the higher capacity
in the near future.
From the Juniper playbook, the QFX10000 Series precisely fits the bill
as it supports 100GbE. The choice is between an eight-slot chassis and
a 2RU fixed-format switch. The 2RU platform provides either 12 or 24
ports of 100GbE, while the 13RU chassis can provide up to 240 x
100GbE via a series of line cards. In that case you could have either the
30C line card supporting 30 x 100GbE, or the 36Q, which supports 12
x 100GbE.
The two options are similar in price, but here the 2RU unit is more
than sufficient for the proposed client solution. It provides 100% scale
even if the client wants a higher capacity bandwidth solution, and it
takes up a smaller amount of space, uses less power, and will require
less cooling. On that basis it would seem strange to pick a larger
chassis when 70% of it would be unused, even if the higher bandwidth
solution is chosen.
Utilizing the QFX10002-72Q would mean our rack design would look
similar to Figure 4.4.

Figure 4.4

Chapter 4: Oversubscription

DC1 Proposed Data Center Rack Design Core, Spine, and Leaf

And using the QFX10002-72Q makes the logical subscription design

look like Figure 4.5.

Figure 4.5

Oversubscription Design of DC1 Core, Spine, and Leaf

The subscription ratio and onward connectivity from the core to the
WAN and its MX Series layer, can be treated slightly differently since
its generally known that 75% of your traffic stays local to the data
center. It seems to be a common traffic profile since servers have
moved to a virtual basis and as the applications are more distributed in
nature, as are their dependencies. As such, you end up with a 75/25%
split in how your traffic traversing your DC, with the 75% local to the
Leaf/Spine/Core, and 25% traversing the WAN.

Day One: Data Center Fundamentals

This means you can provision a higher subscription ratio out to the
WAN router, which in turn means smaller physical links. Again the
choice comes down to 40GbE or 100GbE. While the MX Series
supports both, the cost of those interfaces differs a lot as WAN router
real estate is quite expensive compared to switching products.
In either case you have the option of 40GbE or 100GbE connectivity
from each core switch to the WAN layer. Going 100GbE would be the
better option but it may mean additional costs that the client wasnt
expecting. If thats the case then its easy to provision multi-40GbE
connections instead. In either case, the logical subscription solution will
be similar to that shown in Figure 4.6.

Figure 4.6

Final DC1 Design WAN, Core, Spine, and Leaf

DC2 Design
The second data center presents a different challenge because the client
has stated they would prefer an end-of-row/middle-of-row architecture.
In review of the requirements, DC2 will have five rows of ten racks, the
same as DC1, but with each rack housing five blade chassis per rack
with each chassis connectivity at four 10GbE. The client would also like
a 4:1 or lower subscription ratio, similar to DC1.
So the calculations would be:
Total per a rack = 4 x 5 (number of 10GbE times number of

servers) = 20
Total per a rack is 20 x 10GbE = 200GbE
Per a row - 20 x 10 (number of 10GbE per rack times number of

racks per row) = 2TB (or 200 x 10GbE per a row)

Chapter 4: Oversubscription

As opposed to DC1, all of the connections in DC2 have to be connected to the two end-of-row switches. As stated in an earlier chapter, the
most efficient way would be to provision two chassis, with one at
either end of the rows. Split the connections from each rack by two, so
100 x 10GbE to one end of row and 100 x 10GbE to the other end of
row as shown in Figure 4.7, where each cable from each rack is 10 x 10
GbE.

Figure 4.7

DC2s Design Server to EoR Switch

And from an oversubscription point of view, our first hop from the
servers to the end-of-row switches will be at line rate as shown in
Figure 4.8 below.

Figure 4.8

DC2s Oversubscription - Server to EoR

With this in mind the first point of concern for oversubscription is from
the leaf layer to the spine layer. With 100 x 10GbE per leaf, or 1TB,
and an oversubscription ratio of at least 4:1 would mean 1TB divided
by 4 equals 250GbE of uplink connectivity to the spine layer.

Day One: Data Center Fundamentals

The ideal product for the leaf layer would be the QFX10008 chassis.
It has eight slots, so plenty of room for expansion if additional blade
servers are installed into the racks, and it also has a range of line cards
to suit any combination of connectivity. The main line card of interest
is the 60S-6Q, which has 60 ports of 10GbE plus six extra uplink ports
supporting either 6 x 40GbE, or 2 x 100GbE. You could take the 100 x
10GbE rack connections, divide these over two 60S line cards, and
then use the two 100GbE uplink ports from each line card to connect
to the spine layer. This would drop the oversubscription ratio by half,
as the ratio would be based on 50 x 10GbE ports per line card and 2 x
100GbE uplinks. So, 500 divided by 200 (total bandwidth of incoming
connections / total bandwidth of uplinks) = 2.5 to 1, well within the
clients oversubscription ratio.
So our logical oversubscription design would look like the one illustrated in Figure 4.9.

Figure 4.9

DC2 Oversubscription Ratio

The effect of provisioning 4 x 100GbE per leaf is that the spine layer
has to accommodate 8 x 100GbE per row and with ten rows thats 80
x 100GbE across two spine switches. Given the quantity of 100GbE
thats required, the choice of spine layer device is made for us in the
guise of the QFX10008 chassis.
The most suitable line cards are either the 36Q, which supports 12
ports of 100GbE, or the 30C, which 30 ports of 100GbE Both line
cards also provide breakout options for 40GbE and 10GbE, so you
dont have to provision additional line cards for those connectivity
options. If you remove cost concerns from this design, then the 30C
line card is the obvious choice because you can have two line cards per
spine, providing 60 x 100GbE ports. But when you factor cost concerns into this design, then it might be slightly more economical to
place four 36Q line cards per spine than two 30Cs.
So the physical diagram for DC2 would look like Figure 4.10.

Figure 4.10

Chapter 4: Oversubscription

DC2 Spine and Leaf Design

NOTE

Figure 4.11

Notice that both Figure 4.9 and Figure 4.11 consolidate the number of
layers by using an end-of-row design. The leaf nodes are now in the
end-of-row position and the spine layer now moves up to become what
was the core layer. So the principle of spine and leaf stays the same
regardless of whether you use top-of-rack or end-of-row design.

DC2 The Difference Between DC1 and DC2 Spine and Leaf

Day One: Data Center Fundamentals

If you follow the same principle for connectivity to the WAN layer as
designed for DC1, whereby 75% of the traffic would be local to the
spine and leaf and 25% over the WAN, then the connectivity would
consist of either several 40GbE interfaces from both spines to the
WAN layer, or you could provision 100GbE, but again, thats an
option to offer to the client.
In each case, the logical oversubscription solution will look like the one
shown in Figure 4.12.

Figure 4.12

DC2 to WAN Oversubscription

So now that the designs for DC1 and DC2 are established, lets move
on to logical design and consider the different options available using
the same client requirements as stated at the beginning of this chapter.

Additional Points to Consider

One element not touched upon is support for 1GbE-based servers. In
the scenario for DC1 any one of the servers in the rack design can be
multi-1GbE, but what changes is that the total connection bandwidth
to the top-of-rack switch would be reduced, and as such the uplink
bandwidth from that top-of-rack/leaf to the spine in that row may be
slightly lower. If you run through the calculations again youll find you
may only need a 10GbE uplink connectivity as opposed to 40GbE. In
each case, bear in mind the requirements and the subscription ratio
you need to attain.
From a physical point of view the type of switch you could use to
support 1GbE might also change. The QFX5100 series supports 1GbE
copper natively via a 1GbE copper SFP. If the number of 1GbE connections is greater than the number of 10GbE server connections, and you
suspect that there will be no migration to 10GbE servers, then consider
the EX4300 switch as an option. If, on the other hand, there is likely to
be a migration to 10GbE in the near future, keep the QFX5100 and
use optics to connect the 1GbE requirement. It will save a full switch
swap-out in the future.

Chapter 5
Fabric Architecture

In this chapter we move from the physical applications of data

center design discussed in Chapter 4 to the different logical
architectures that Juniper provides. When discussing logical
design, this Day One book refers to the IP addressing structure
and the Layer 2 and Layer 3 transport mechanisms essentially
the control plane and data plane of the design. Juniper provides
you with several options, from single switch management with
multiple transport mechanisms, to virtualizing the control plane
across multiple switches creating a fabric with a single point of
management.
First of all, why do we need a logical network? This question
arises every time there is an increased need to scale the network in
order to support new technologies such as virtualization.
There are protocol and control mechanisms that limit the disastrous effects of a topology loops in the network. Spanning Tree
Protocol (STP) has been the primary solution to this problem
because it provides a loop-free Layer 2 environment. Even though
STP has gone through a number of enhancements and extensions,
and even though it scales to very large network environments, it
still only provides a single active path from one device to another,
regardless of how many actual connections might exist in the
network. So although STP is a robust and scalable solution to

Day One: Data Center Fundamentals

redundancy in a Layer 2 network, the single logical link creates two

problems: at least half of the available system bandwidth is off-limits
to data traffic, and when network topology changes occur they can
take longer to resolve than most applications will accept without loss
of traffic. Rapid Spanning Tree Protocol (RSTP) reduces the overhead
of the rediscovery process, and it allows a Layer 2 network to reconverge faster, but the delay is still high.
Link aggregation (IEEE 802.3ad) solves some of these problems by
enabling the use of more than one link connection between switches.
All physical connections are considered one logical connection. The
problem with standard link aggregation is that the connections are
point-to-point.
So in order to remove STP and its somewhat limited functionality,
Juniper, like other vendors, came up with a series of different logical
solutions for the network that eliminate the reconvergence issue and
utilize 100% of all available bandwidth.
Junipers architectures are based on the principle of a common physical
platform centered on the QFX Series, but supporting different logical
solutions. This approach provides you with the flexibility to support
different options and fit the scenario you need, but it also means that if
your requirements change you can adapt both the physical and logical
structure without costly repurchases.
Juniper offers five logical architectures: two of them are open standards (1 and 5), and three are Juniper-based solutions (2, 3, and 4):
1. Traditional MC-LAG (multichassis link aggregation group)
2. Virtual Chassis
3. Virtual Chassis Fabric
4. Junos Fusion
5. IP clos fabrics
Lets use the requirements for DC1 and DC2 architectures outlined in
Chapter 4 as the basis of a fabric architecture discussion, so each
logical offering can be imposed over the top of the physical architecture. Along the way this chapter outlines the details of each solution,
its merits, and which may be more appropriate to which design. Lets
use the solutions created in Chapter 4, minus the WAN layer, and use
Figure 5.1 for this chapters base physical architecture.

Figure 5.1

Chapter 5: Fabric Architecture

Chapter 5 Base Physical Architecture

Traditional MC-LAG
Our starting point is MC-LAG, which enable a client device to form a
logical link aggregation group (LAG) interface between two MC-LAG
peers. An MC-LAG provides redundancy and load balancing between
the two MC-LAG peers, multi-homing support, and a loop-free Layer
2 network without running STP.
In Figure 5.2, on one end of an MC-LAG is an MC-LAG client device,
such as a server or switch, which has one or more physical links in a
LAG. This client device uses the link as a LAG and traffic is distributed
across all links in the LAG. On the other side of the MC-LAG, there
are two MC-LAG peers. Each of the MC-LAG peers has one or more
physical links connected to a single client device.

Figure 5.2

Basic MC-LAG

Day One: Data Center Fundamentals

The MC-LAG peers use Inter-Chassis Control Protocol (ICCP) to

exchange control information and coordinate with each other to
ensure that data traffic is forwarded properly. ICCP replicates control
traffic and forwarding states across the MC-LAG peers and communicates the operational state of the MC-LAG members. Because ICCP
uses Transmission Control Protocol/ Internet Protocol (TCP/IP) to
communicate between the peers, the two peers must be connected to
each other. ICCP messages exchange MC-LAG configuration parameters and ensure that both peers use the correct Link Aggregation
Control Protocol (LACP) parameters.
The interchassis control link (ICL), also known as the interchassis
control link-protection link (ICL-PL), is used to forward data traffic
across the MC-LAG peers. This link provides redundancy when a link
failure (for example, an MC-LAG trunk failure) occurs on one of the
active links. The ICL can be a single physical Ethernet interface or an
aggregated Ethernet interface. You can configure only one ICL between the two MC-LAG peers, although you can configure multiple
MC-LAGs between them.
LACP is a subcomponent of the IEEE 802.3ad standard and is used to
discover multiple links from a client device connected to an MC-LAG
peer. LACP must be configured on both MC-LAG peers for an MCLAG to work correctly.
Figures 5.3 and 5.4 show that MC-LAG can be configured in active/
standby mode, where only one device actively forwards traffic, or in
active/active mode, in which both devices actively forward traffic. In
active/standby mode, only one of the MC-LAG peers is active at any
given time. The other MC-LAG peer is in backup (standby) mode. The
active MC-LAG peer uses the LACP to advertise to client devices that
its child link is available for forwarding data traffic.
In active/active mode, Figure 5.4, you can see that all member links are
active on the MC-LAG, as opposed to Figure 5.3, which shows active/
standby mode. In active/active mode, media access control (MAC)
addresses learned on one MC-LAG peer are propagated to the other
MC-LAG peer.
In most cases active/active is the preferred mode.
Traffic is load-balanced in active/active mode, resulting in 100%

use of all links.

Chapter 5: Fabric Architecture

Figure 5.3

MC-LAG Active / Standby Mode

Figure 5.4

MC-LAG Active / Active Mode

Convergence is faster in active/active mode than in active/standby

mode. In active/active mode, information is exchanged between

devices during operations. After a failure, the operational switch
or router does not need to relearn any routes and continues to
forward traffic.
Active/active mode enables you to configure Layer 3 protocols on

integrated routing and bridging (IRB) interfaces, providing a

hybrid Layer 2 and Layer 3 environment.

Day One: Data Center Fundamentals

But how does this relate to the DC1 and DC2 scenarios? The quick
answer is: it depends. Recall that in earlier chapters, our two options
were spine and leaf in the form of a top-of-rack and a spine and leaf in
an end-of-row topology.
MC-LAG relates to Layer 2-based solutions and as such if your proposed network is going to have Layer 2 VLANs present at both the leaf
layer and the spine layer, in both solutions, then you are going to need
MC-LAG at either the spine layer for DC1 and either the spine or leaf
layer for DC2. As Figure 5.5 shows, the spine layer would have to be
interconnected to each other for a MC-LAG to work. It also means your
MC-LAG setup for every Layer 2 domain is present at a central point in
the network, as opposed to distributed between multiple leaf devices,
which is possible but can be messy.

Figure 5.5

MC-LAG Architecture for DC1

With the DC2 logical architecture you have the option to implement one
step closer to the server layer at the leaf layer as shown in Figure 5.6.

Figure 5.6

MC-LAG Architecture for DC2

Chapter 5: Fabric Architecture

Now the question is, should you implement MC-LAG in either of these
scenarios? I would suggest not. Solutions like VCF, Junos Fusion, and a
Layer 3 IP Clos with a Layer 2 overlay would be far better solutions,
and they scale considerably further and with less management overhead.
Where MC-LAG setups do become relevant is when you dont want a
proprietary fabric or an IP fabric but need a failover architecture for a
small deployment of QFX Series in a data center. In this instance,
MC-LAG would be appropriate and is fully supported across all QFX
Series platforms.
MORE? For more detailed information on MC-LAG and the QFX Series refer to

the following link, which outlines configuration examples and best

practices: http://www.juniper.net/techpubs/en_US/junos/informationproducts/pathway-pages/mc-lag/multichassis-link-aggregation-groups.
html.

Virtual Chassis
Juniper released Virtual Chassis on the EX4200 Series Ethernet Switch
in 2009 as an alternative to the stacking technologies that were, and are
still, present. The basic principle in Virtual Chassis is to take the benefits
of a single control plane in a chassis and virtualize it over a series of
distributed switches interconnected via a cable backplane. It allows the
management of up to ten switches to be simplified via a single management interface, so you have a common Junos OS version across all
switches, a single configuration file, and an intuitive chassis-like slot and
module/port interface numbering scheme.
The design is further simplified through a single control plane and the
ability to aggregate interfaces across Virtual Chassis switch members as
you would with a normal chassis.
Switches within a Virtual Chassis are assigned one of three roles; master
RE (Router Engine), backup RE, and line card. As the name implies, the
master RE is the main routing engine for the Virtual Chassis. It acts as
the main point of configuration and holds the main routing table, which
is copied to each Virtual Chassis member to provide local, or distributed, switching and routing.
The backup provides a backup to the master RE and it is ready to take
over mastership of the Virtual Chassis in the event of a failure of the
master. The backup maintains a copy of the switching and routing tables
and when an update to the master is enacted, then the backup RE
automatically updates its tables. Think of it as an active/active sync.

Day One: Data Center Fundamentals

Figure 5.7

Virtual Chassis Switch Roles

All other switches are classed as line cards and in the event a master RE
goes down and the backup RE takes over mastership, then a switch
member classed as a line card automatically becomes the backup RE. In
reality, when the old master RE has resolved its issue, it should come
back as either the master or new backup. You have the flexibility to
select the best option depending on your needs.
The special sauce that makes all this possible is Junipers proprietary
protocol called, unsurprisingly, Virtual Chassis Control Protocol
(VCCP). Its based on Intermediate System-to-Intermediate System
(IS-IS), which is a link state protocol. Its based on the principle of
discovering, maintaining, and then flooding topology information on
the shortest paths across the Virtual Chassis to all members with the
most efficient routes on a flow-by-flow basis. VCCP is not user-configurable, and operates automatically, on both the rear-panel and frontfacing Virtual Chassis ports.
So how does Virtual Chassis apply to this books data center solution?
Besides the EX Series lineage and tradition, Virtual Chassis is also
supported on the QFX5100 Series and that gives you the option to
contrast a Virtual Chassis for both the top-of-rack and end-of-row
data centers.
In fact, the ability to mix and match different switch types in a single
Virtual Chassis is a fundamental feature and Virtual Chassis supports
these different Juniper Networks EX Series switching platforms:
EX4200
EX4550
EX4300
EX4600
QFX5100

Chapter 5: Fabric Architecture

The EX4200 and EX4550 can be part of the same Virtual Chassis, and
the EX4300 and EX4600 can be part of the same Virtual Chassis, but
you can also mix the EX4300 and the QFX5100 in the same Virtual
Chassis, which could be an option in our DC1 and DC2 to support
100/1000Mbps interfaces as well as 10GbE in the same virtual switch.
Lets investigate the design option as shown in Figure 5.8.

Figure 5.8

Virtual Chassis Architecture for DC1

For DC1, Virtual Chasis could run the entire top-of-rack switches on a
rack-by-rack basis. This would give a common fabric at the rack layer,
while allowing traffic to switch locally without having to traverse the
spine layer, as all members of the Virtual Chasis to be interconnected
(hence the Virtual Chasis flow in Figure 5.8). The oversubscription
ratio would stay the same and even be reduced slightly due to the local
switching at the rack layer.
But the cabling costs go up and you have to provision additional cables
between each of the switches. Also, traffic isnt using the most efficient
route to get to its destination, as traffic from top-of-rack 1 would need
to pass through all of the other switches to get to top-of-rack10. This is
due to Virtual Chassis assigning a higher weighting in its calculation
than sending the traffic externally up to the spine and down to top-ofrack 10, even though thats the more efficient route. The other negative
factor is scale. While the requirements are for ten switches over ten
racks, what if you need to introduce more switches, or create a secondary Virtual Chassis in the same racks? Either situation would just add
to the complexity of the solution.
This is why Virtual Chassis Fabric was introduced, to allow you to
take the best elements of Virtual Chassis and institute better scale and
traffic distribution. Its discussed in the next section.

Day One: Data Center Fundamentals

For DC2, Virtual Chassis only works well when you introduce a
stack-based solution into the two end-of-row cabinets as opposed to a
chassis, as shown in Figure 5.9. The limitations still exist, however,
which means that while a small implementation of Virtual Chassis
could work, you are still going to be limited by the backplane capacity.
In a Virtual Chassis this is comprised of a number of Virtual Chassis
port (VCP) links and could lead to a higher internal oversubscription
because your traffic distribution wont take the most efficient path,
because traffic potentially still needs to pass through multiple switches
to reach a destination.

Figure 5.9

Virtual Chassis Architecture for DC2

So while Virtual Chassis is an option, I would suggest that Virtual

Chassis favors small data center deployment solutions as opposed to
anything at scale. Remember, its a great campus technology and it does
offer some excellent options, but there are other DC fabric designs that
provide more robust solutions, as is outlined in the next section.

Virtual Chassis Fabric

Virtual Chassis Fabric takes the principles of Virtual Chassis with a
single control plane over multiple switches, but goes much larger in
scale, supports multiple higher port types (1/10/40GbE), and introduces
the IP Clos topology that simplifies the fabric.
What is a fabric? A data center fabric is a system of switches and servers
and all the interconnections between them that visually resemble a
fabric. A data center fabric allows for a simple flattened architecture
where any connecting device is only a single hop away.

Chapter 5: Fabric Architecture

The flattened architecture of fabrics is key to their agility. Fabrics are

comprised of two elements, a leaf and a spine, which are configured in
a hub and spoke design. The leaf node provides the connectivity to
servers while the spine provides the aggregation point for various
leaves as shown in Figure 5.10.

Figure 5.10

Virtual Chassis Fabric Leaf and Spine

Virtual Chassis versus Virtual Chassis Fabric Comparisons

In a spine and leaf topology, the total number of distinct paths between
two leaf devices is equal to the number of spine devices. In Figure 5.11,
each leaf device has four connections to four spine switches, and as
such, it has four paths to every other leaf switch. If any link goes down,
it will end up having fewer paths. But as long as there is more than one
such path, traffic will be distributed on all available paths.
As a result, it provides a distinct advantage over Virtual Chassis in that
every location within the fabric is only two hops away, as opposed to
Virtual Chassis where a destination could be up to nine hops away.
With Virtual Chassis Fabrics increase in scale you can now move past
the limitations of Virtual Chassis, which only allows up to 10 members
to be virtualized in to a single solution. Virtual Chassis Fabric allows
up to 20 switches, with either two or four spines and 16 or 18 leaves,
allowing you to build a single virtual switch to support up to 864 x
10GbE ports per Virtual Chassis Fabric if you use 48 port leaf switches
and 1,728 x 10GbE if you were to use 96 port leaf switches.
The principle of a master and backup in an active/passive routing
engine stays the same as outlined in Virtual Chassis but generally we
now elect the spine layer switches as the master and backup utilizing a

Day One: Data Center Fundamentals

single virtual MAC and IP gateway. All other switches in the Virtual
Chassis Fabric become line cards of the master. Theres a single point
of CLI through the master, and theres a distributed forwarding plane
so every switch has a copy of the both the Layer 2 and 3 tables.

Figure 5.11

Master, Backup, and Line Card Placement

With Virtual Chassis Fabric you also gain the option of plug and play
implementation. In the existing Virtual Chassis, you had to designate
the master and backup, and then you could pre-prevision the other
members of the Virtual Chassis though configuring the serial numbers
into the master so the master knew what line cards would be members.
That option still exists, but you also have the option to pre-prevision
just the master and backup at the spine layer and any other nodes that
are directly connected to any of the spine nodes are allowed to join the
Virtual Chassis automatically without pre-provisioning their serial
numbers. In this mode, leaf nodes can be brought into the Virtual
Chassis Fabric without any user intervention their configuration,
VCP port conversions and dynamic software update will be handled
automatically through the use of the open standard LLDP (Link Layer
Discovery Protocol).

Chapter 5: Fabric Architecture

With the higher bandwidth achieved through Virtual Chassis Fabric,

the way traffic paths are calculated has changed as well. In a Virtual
Chassis system the shortest path calculation strictly followed the
Dijkstra algorithm, which evaluated both the number of hops and link
bandwidth between members of the Virtual Chassis. But, even when
there were multiple equal cost paths between two devices, only one
path would be used for data traffic.
In Virtual Chassis Fabric, the VCCP protocol is enhanced to discover
all distinct paths. When multiple paths are discovered, traffic is now
load-balanced across these multiple paths and these paths are formed
into virtual trunks or aggregated links. Virtual Chassis Fabric has two
types of trunks: next hop trunks (NH trunks) and remote destination
trunks (RD trunks). NH trunks are trunks that are for devices that are
directly connected to each other, such as between a leaf and a spine,
and when there are multiple links between two physical devices, a VCP
trunk is created to simplify topology.
To achieve load balancing across multiple paths, VCCP was enhanced
further to create a trunk to each remote destination node. So if traffic
from one leaf needs to get to another leaf via the spine layer, the VCCP
algorithm also takes into account the paths to the remote destination
when calculating the best path and forming a virtual trunk to that
destination.
To remove the possibly of congestion, the VCCP protocol also takes
into account the bandwidth of each link, and implements a fair weight
algorithm so each node or switch in the Virtual Chassis Fabric can
spray traffic across the fabric on a flow-by-flow basis VCCP considers the bandwidth on each link, both physically and to its remote
destination, and automatically adjusting traffic on a link-by-link basis
to avoid congestion and deliver packets in order. The protocol that
does this is Adaptive Flowlet Splicing.

Virtual Chassis Fabric in DC1 and DC2

So how does Virtual Chassis Fabric help us in our two DC scenarios?
For DC1, which follows a top-of-rack topology, Virtual Chassis
Fabric is a perfect fit. Our top-of-rack switches become leafs with
connectivity up to the spine layer. Our two spine switches become
master RE and backup RE and all of our leaves become line cards as
shown in Figure 5.12.

Day One: Data Center Fundamentals

Figure 5.12

Virtual Chassis Fabric in DC1

As you can see in Figure 5.12, the links between leaf and spine now
become Virtual Chassis Ports and allow the Virtual Chassis Fabric to
become a single fabric over which traffic will be distributed. Northbound connectivity from the Virtual Chassis Fabric to the core is via
the spine layer advertised with the single IP gateway for the Virtual
Chassis Fabric. The purpose of the core layer in this design is to
provide a point of aggregation for the Virtual Chassis Fabric to
connect to the other Virtual Chassis Fabrics present in the other rows.
You can also LAG the links up to the core to allow traffic to be distributed to both spines as shown in Figure 5.13.

Figure 5.13

Virtual Chassis Fabric in DC1 with Core Layer

Chapter 5: Fabric Architecture

This design can then be replicated over the other rows of racks allowing what were 50 racks of switches to be virtualized into five IP
addresses and five points of management for the whole of DC1.
Virtual Chassis Fabric also provides full support for SDN-controller
based solutions integrated with virtual servers which is covered in
Chapter 9.
For DC2, which uses an end-of-row based solution, Virtual Chassis
Fabric becomes a little more challenging. With all of the switches
located in single racks at each end of the row you would need to
implement a spine in each row and then interconnect each leaf across
the rows to reach each spine (as shown in Figure 5.14) to allow the
Virtual Chassis Fabric to form. As noted earlier, for a Virtual Chassis
Fabric to form every leaf needs to be directly connected to each spine
layer. This would increase cabling issues and costs. The other option
would be to implement a Virtual Chassis Fabric-per-end-of-row, with
two spines and rest of the switches as leafs, but stacked. That would
provide a single point of management the same as a chassis would
provide, but would require a major increase in cabling as you would
have to follow the same principles of leaf and spine connectivity but
within a single rack. In addition, the oversubscription ratio is different
for DC2 and as such would need to increase the cabling at 40GbE to
make a Virtual Chassis Fabric work.

Figure 5.14

DC2 Virtual Chassis Fabric Stretch Over a Row of Racks

Day One: Data Center Fundamentals

This is why you need to work through the pros and cons for each
scenario. Virtual Chassis Fabric works really well for a top-of-rack
based solution, but for end-of-row it becomes a little more problematic
and I suspect it would still be better to position a chassis for the end-ofrow solution as opposed to individual switches.
Our next architecture takes many of the principles of Virtual Chassis
and Virtual Chassis Fabric and enlarges the scale of the number of
switches you can virtualize.

Junos Fusion
Junos Fusion is a new technology from Juniper, which, at its most basic,
takes the elements of Virtual Chassis Fabric, such as the single management domain, the removal of spanning tree, and the full bandwidth use,
and expands the support up to 128 switches per virtualized domain
while simplifying the management of all those switches through the use
of a new software implementation.
The Junos Fusion architecture is comprised of Aggregation Devices
(QFX10000 switches) and Satellite Devices (EX43000 and/or
QFX5000 switches). Satellite devices directly connect to the aggregation device in a hub and spoke topology as shown in Figure 5.15.

Figure 5.15

Junos Fusion Setup

Chapter 5: Fabric Architecture

The key characteristic of Junos Fusion is that satellite devices can be

managed either by a single aggregation device or a pair of aggregation
devices. Think of the satellite devices as line cards of the aggregation
device but with the feature set of the aggregation devices. This provides
customers with simplicity of operations, enabling them to manage a
large data center from just a pair of switches (aggregation devices)
without having to actively manage any of the top-of-rack switches,
since they are managed centrally from the aggregation devices.

How Junos Fusion Compares to Virtual Chassis Fabric?

The key difference between Junos Fusion and Virtual Chassis Fabric is
that Virtual Chassis Fabric is comprised entirely of switches that use
Broadcom PFEs. As a result, they use a fabric protocol called HiGig to
communicate within the Virtual Chassis Fabric architecture. In Junos
Fusion, since the architecture consists of combining systems with both
Broadcom PFEs and Junipers own Q5 ASIC (on the QFX10000), the
HiGig protocol isnt used. Instead, Junos Fusion uses the open standard IEEE 802.1BR bridge port extension protocol to communicate
between various devices, an open, standards-based mechanism to
remotely manage top-of-rack switches or satellites devices.
Another key difference is scale. While Virtual Chassis Fabric scales to
about 20 racks, Junos Fusion scales to a much higher 128 racks, or 128
satellites.
Some of the key attributes of aggregation devices within the Junos
Fusion architecture are:
One or more switches that manage satellite devices (top-of-rack

switches) remotely
Configuration, software image management, statistics polling

from satellite devices

The system supports an option to replicate the configuration

between aggregation devices. Satellite devices do not need to

keep a local configuration copies.
Automated discovery and provisioning of satellite devices
In-band connectivity for management of satellite devices
Junos OS support with central upgrade process of the satellite

devices. Junos Fusion has been designed in such a way that all
components are loosely coupled. This allows us to introduce the
notion of software upgrade groups. Devices within the same
SUG, or software upgrade group, need to run the same OS. A
given Junos Fusion system can have as many SUGs as the opera-

Day One: Data Center Fundamentals

tor prefers, and the OS across different SUGs does not have to be
the same. This enables operational simplicity and ease of maintenance.
Supported Aggregation Device: QFX10002-72Q, 36Q,

QFX10008, and QFX10016

Cascade ports (satellite ports) can be used as regular L2 or L3

ports for connecting any type of devices such as firewalls and

load-balancers
And some of the key attributes of satellite devices within the Junos
Fusion architecture are:
Standard top-of-rack switches managed remotely by one or more

aggregation devices
No local management required
Can be single or dual-homed to aggregation devices
Satellite devices run Windriver Yocto Linux or Linux Forwarding

Engine OS (LFE OS) using considerably less space than a normal

Junos software image, thus making the upgrade process faster
Supported Satellite Device: EX4300 and QFX5100 switches

One attribute not highlighted above is licensing. Because the Junos

Fusion satellite devices essentially inherent the features of the aggregation device, the protocols supported on the aggregation device are now
present on the satellite, which means you now only place the license at
the aggregation. So if you need OSPF at the server layer, then once the
satellite is plugged in and registered, it will have full OSPF but without
the need for the additional license.

Junos Fusion in DC1 and DC2

So how does Junos Fusion help with the two DC scenarios in this
chapter? For DC 1 it can change the architecture of the design as we
now can remove the spine layer and allow the top-of-rack switches to
connect directly to the core layer, due to the fact that Junos Fusion
doesnt support a tiered layer approach. This would allow you to
reduce the number of devices in the solution by removing a layer of the
network and you simplify the management domain as all switches in
all rows are now under the same single point of management.
Note that there are changes whose effects can be significant for the
client. The first change is that the new spine/core layer will need to

Chapter 5: Fabric Architecture

increase to accommodate the number of 40GbE connections that are

now coming firstly from each top-of-rack switch. This has the effect of
increasing the cost of the solution as core devices are swapped for
larger chassis-based switches and you increase the number of 40GbE
line cards. While increasing the cost of the core/spine there is also the
increase in cost of fiber connectivity as each top-of-rack now connects
to a central core switch.
NOTE

Junos Fusion comes with a lot of failure protections built-in, and, for
the sake of brevity, they are not covered here, but Junos Fusion is a
active/active control plane across the data center, and as such the
potential for a control plane failure affecting the whole data center
should be a concern you need to understand so you can incorporate its
management into your best practices.
Is Junos Fusion an option for DC1? Yes, if you want to simplify the
management domain to cover a large number of switches while
providing an easy point of configuration and management.

Figure 5.16

Junos Fusion in VC1

As for DC2, the potential issue with Junos Fusion is the same as
implementing Virtual Chassis Fabric. Youre dependent on 1RU QFX
devices to do the 10GbE termination in both solutions. This would
mean implementing stacks of 1RU switches and connecting those
satellite devices back to the core/aggregation layer. The benefit of the
chassis solution is that local switching is done via the backplane on the
chassis, as opposed to pushing the traffic up to the core and potentially
back down to the leaf layer.

Day One: Data Center Fundamentals

Again, this is why Juniper provides you with options, because one
solution does not always fit all scenarios, and while Junos Fusion
works well for DC1 more than DC2, maybe an IP fabric as outlined in
Chapter 6 could provide a solution for both.
MORE? For more detailed information about Junos Fusion and data centers,

see the Juniper/OReilly book, The QFX10000 Series, by Douglas

Hanks (OReilly Media, 2016), at http://www.juniper.net/books.

Chapter 6
IP Fabrics and BGP

Why give IP fabrics its own chapter?

Quite simply, its extremely flexible with respect to the different
topologies that it can support, it brings proven carrier-class Internet
scaling architecture into the data center with the ability to scale from
a few switches to several thousand while keeping the same underlying principle, and it provides a stable foundation for IP connectivity
so you can run virtual overlay networks.
The other reason is that Virtual Chassis Fabric and Junos Fusion are
fairly straightforward, and IP fabrics are less plug and play, so they
need a little planning in their design to implement, and subsequently
need a little more space to explain all of the features and their
relevance.

IP Fabrics
IP Fabrics have been popular of late because they provide the best
support for virtual servers and their applications, and they offer
the ability for those applications to talk to each other at scale via
Layer 2. While the virtual server world has been moving at breakneck speed to provide speed and agility, the networking world has
been a little slow to respond.
Up until a few years ago, most data center networks were based on
the traditional three-tiered approach that had been copied from
campus network design. While its fine when most of the traffic has a
north/south profile like a campus network, its not really suitable
when the majority of virtual applications and associated workloads

Day One: Data Center Fundamentals

need constant east/west communication. East-west traffic is traffic that

occurs within the data center and it dominates in nearly all cases of
cloud computing, virtualization, and big data. It is one of the driving
forces behind the development of data center fabrics.
Multi-tier network data center architectures were optimized for the
older north-south traffic, direct traffic from the data center to the end
user, not east-west traffic. So in the old design, Layer 2 VLANs would
traverse from the server access switch to the aggregation layer and then
potentially up to the core layer as shown in Figure 6.1.

Figure 6.1

Traditional Multi-tier Approach

As you can see in Figure 6.1, this design introduces a fair amount of
latency, as the packets need to traverse several hops in order to reach
their destination. The more hops required between the different
dependent applications, the more those applications are subjected to
additional latency, contributing to unpredictable performance and user
access to those applications.
This is why fabrics have become so popular; they reduce latency by
flattening the architecture so any server node can connect to any other
server node with a maximum of two hops, as shown in Figure 6.2.
Data center fabric architectures typically use only two tiers of switches
as opposed to older data centers that implemented multiple tiered
network architectures.

Figure 6.2

Chapter 6: IP Fabrics and BGP

IP Fabrics Two Tier Approach

Traffic can be transmitted between server nodes in almost all data

center fabric architectures by traversing a set number of switches,
which results in extreme efficiency and low latency. The fabric consists
of multiple direct paths of high bandwidth, thus removing any potential transmission slowdown caused by network bottlenecks.
Juniper led the innovation of this principle with the release of Q-Fabric
in 2010. The idea was to provide single-hop connectivity between
servers across a high-speed fabric but with an out-of-band control
plane that was connected to every switch device.
An IP Fabric evolves the principles of Q-Fabric to its next logical step
by implementing the control plane in software as opposed to a physical
out-of-band network and provides the provision for overlay. The
majority of IP Fabric architectures are represented by a spine and leaf
design in which the fabric mesh incorporates devices on the edge (the
leaves) of the fabric and switches on the spine. The spine and leaf
model allows for simplified expansion limited only by the number of
supported devices and their ports, as shown in Figure 6.3 and as we
have covered in previous chapters.

Day One: Data Center Fundamentals

Figure 6.3

IP Fabric Architecture

You can see in Figure 6.3 that the spine layer is made up of four
switches. Each leaf has four uplinks, one to each spine. The maximum
number of leaves supported in this topology is dictated by the maximum number of ports per a spine. So if our spine switch supports 40 x
40GbE connections, the maximum number of leaf devices would be 40
(though we should allow for onward connectivity at a similar rate to
the uplink connectivity, so 36 would be about right).

Clos
Where did the spine and leaf architecture come from? Its based on the
principles of a Clos network. Clos networks are named after Bell Labs
researcher Charles Clos, who proposed the model in 1952 as a way to
overcome the performance and cost-related challenges of electro-mechanical switches then used in telephone networks. Clos used mathematical theory to prove that achieving non-blocking performance in a
switching array (now known as a fabric) was possible if the switches
were organized in a hierarchy.
The design Charles Clos came up with was what he classed as a three
stage Clos. A three stage clos is comprised of an ingress, a middle, and
an egress, as shown in Figure 6.4.

Chapter 6: IP Fabrics and BGP

Figure 6.4

Three Stage Clos (Ingress, Middle, and Egress)

As Figure 6.4 shows, the number of sources which feed into each of the
ingress switches is equal to the number of connections connecting to
the middle switches and is equal to the number of connections connecting to the egress switches. This provides a non-blocking architecture
with no oversubscription. If you now re-name the ingress, middle, and
egress, to leaf and spine, you get a Three Stage Clos as shown in Figure
6.5:

Figure 6.5

Three Stage Clos (Leaf, Spine, and Leaf)

Now if you fold this topology in half and turn it on its side, youll end
up with the same spine and leaf architecture as you have seen previously and as outlined in Figure 6.6.

Day One: Data Center Fundamentals

Figure 6.6

Folded Three Stage Clos (Leaf, Spine, and Leaf)

The number of incoming connections is still equal to the number of

connections between leaves and the spines and there is still a folded
three stage Clos. As discussed later in this chapter, traffic can now be
distributed over all available links with no fear of oversubscription.
As more connections are introduced to the leaf layer, our oversubscription ratio increases and you can increase the size of the interconnects to
the spines to suit the bandwidth connecting in.
Besides its support for overlay technologies the other benefit of this
design is that it offers finely grained failure domains because every
switch in the leaf layer is connected to every switch in the spine layer.
Thus a failure in one switch in either of the layers does not bring down
the entire fabric. As well discuss in the next section, you can utilize
open standard protocols so you can interoperate with other vendors,
which in itself will mean lower CapEx, OpEx, and simpler management.
How does traffic traverse the fabric? Quite simply, by using Layer 3.

BGP
Layer 3 acts as the control plane in allowing routing information to be
distributed to all switches in the fabric. But as you are no doubt aware,
there are several different Layer 3 protocols you can choose from. Best
practice would argue in favor of any one of the three main open standard protocols: OSFP, IS-IS, or BGP would do the job. Each protocol
can essentially advertise routing prefixes, but each protocol varies in
terms of scale and features.
Open Shortest Path First (OSPF) and IS-IS use a flooding technique to
send updates and other routing information. Creating areas can help
limit the amount of flooding, but then you start to lose the benefits of an
SPF routing protocol. On the other hand, Border Gateway Protocol
(BGP) was created from the ground up to support a large number of
prefixes and peering points. The Internet and most service providers are
the best examples of BGP as a control plane at scale.

Chapter 6: IP Fabrics and BGP

Support for traffic engineering is another requirement that would be

useful if you need to steer traffic around a specific spine switch that
requires a software upgrade. The limitations with IS-IS and OSFP start
to become apparent as these protocols have limited traffic engineering
and tagging capabilities, but BGP support for traffic engineering is
fundamental to BGPs design and has several features to allowing the
tagging of packets:
Local preference (a metric for internal neighbors to reach

external paths default is 100)

Multiple exit discriminator (MED) (a metric for external neigh-

bors to reach the AS default is 0)

And, extended communities (Routing Tagging)

Its hard to find a reason why you wouldnt use BGP, unless you have
had little cause up until now to use BGP in any part of your network.
From a campus point of view, many people would consider BGP as a
service provider-based protocol, but the fear of learning this complex
protocol and how to configure it is being surpassed by automation
tools such as OpenClos, which automates the whole process: not only
the BGP configuration for each switch, but also the push of the configuration to the switch and bringing the whole fabric up.
Using an automation tool such as OpenClos will still require you to
have a basic understanding of BGP. So lets conduct a basic review of
BGP basics and BGP design.

BGP Basics
BBGP is an exterior gateway protocol (EGP) used to exchange routing
information among routers in different autonomous systems (AS).
An AS is a group of routers that are under a single technical administration. BGP routing information includes the complete route to each
destination and it uses the routing information to maintain a database
of network reachability information, which it exchanges with other
BGP systems. BGP uses the network reachability information to
construct a graph of AS connectivity, which enables BGP to remove
routing loops and enforce policy decisions at the AS level.
BGP uses TCP as its transport protocol, using port179 for establishing
connections. The reliable transport protocol eliminates the need for
BGP to implement update fragmentation, retransmission, acknowledgment, and sequencing.
An AS is a set of routers or switches that are under a single technical
administration and normally use a single interior gateway protocol
and a common set of metrics to propagate routing information within

Day One: Data Center Fundamentals

the set of routers. To other ASs, the AS appears to have a single routing
plan and presents a consistent picture of what destinations are reachable through it.
The route to each destination is called the AS path, and the additional
route information is included in path attributes. BGP uses the AS path
and the path attributes to completely determine the network topology.
Once BGP understands the topology, it can detect and eliminate
routing loops and select among groups of routes to enforce administrative preferences and routing policy decisions.
BGP supports two types of exchanges of routing information: exchanges between different ASs and exchanges within a single AS.
When used among ASs, BGP is called external BGP (EBGP). When
used within an AS, BGP is called internal BGP (IBGP), which leads us
quite nicely in to the next section.

BGP Design
The first question posed by using BGP is which variant to use: IBGP or
EBGP. The difference between the two may seem quite small, but those
minor differences can lead to major changes in your DC implementation. The first big difference between the two is the way they use
autonomous systems or ASs.
In IBGP, all switches in the spine and leaf sit under a single AS, as
shown here in Figure 6.7. And in EBGP, each switch within the spine
and leaf has its own AS, as shown in Figure 6.8.

Figure 6.7

IBGP Example

Chapter 6: IP Fabrics and BGP

The next issue is how each switch in the fabric distributes prefixes or
routes.
IBGP

In BGP you need a full mesh of connected nodes or switches to allow

the full distribution of prefixes so each node in the fabric knows of the
other, meaning that each node needs to be logically connected to every
other node to create a full mesh. But you still need a method of distributing routes between these nodes so they know how to route traffic
efficiently. To get address this issue you can use two IBGP options
called confederations or route reflectors.
A confederation is simply a separate internal AS that sits inside your
existing single AS. The idea is to break up the internal AS into smaller
parts and tie them together with EBGP. Each internal group will still
have a full mesh, but it can be as finely grained as youd like. But in this
scenario, using confederations could be a configuration nightmare
especially when theres an easier alternative: route reflectors.
A route reflector is an IBGP router that will re-advertise routes to other
IBGP routers. This works by creating clusters of IBGP routers, and
connecting them with a reflector. So for the DC solution it means the
spines become reflectors and the leaves become clients of those reflectors, or route reflector clients.
A reflector will not send every single route; it only sends the best paths
to its peers. This presents a problem when you have two or four spines
and potentially four routes for traffic to take, from the leaf to the spine.
To remove this issue you can enable the Add-Path feature on BGP
route reflectors so they can advertise four routes of equal measure to
the leaf, and traffic can be distributed over all four using equal-cost
multi-path (ECMP).
ECMP is a solution whereby next-hop packet forwarding to a single
destination can occur over multiple best paths that tie for top place in
routing metric calculations, thus allowing load balancing of packets
over those multiple equal paths. So, in normal speak, there are two
spine switches and a leaf is connected to both. Normally, IBGPs route
calculation would favor one over the other, but you want traffic to be
distributed over both, so ECMP would be the normal item to implement. But IBGP doesnt support ECMP as a standard, so you must
enable the add-path feature we mentioned, which allows the best route
to be advertised over both links, and then you can use ECMP to
distribute the traffic over those two links.

Day One: Data Center Fundamentals

Something to consider when choosing between the two methods is that

you have a central point of route management for the fabric, and over
time it can become quite large as its holds all the routes for all of the
switches in your fabric. As you expand the leaf layer, the route table
will get larger and larger and can become quite difficult to manage.
In summary, the spine switches must support BGP route reflection as
well as the BGP Add Path feature to meet all of the IP fabric requirements, and your route reflectors may over time become difficult to
manage as the configuration gets bigger and bigger but this allows
you to manage the entire IP fabric with a single autonomous system
number as opposed to eBGP where you would have to consider the
number of ASs you may use.
EBGP

EBGP is easy in comparison to IBGP, because theres no need for BGP

route reflection or BGP Add Path to be enabled as ECMP will distribute traffic over all the best paths available and advertised between
peers.

Figure 6.8

EBGP Example

The only issue is the number of ASs you might consume with the IP
fabric. Each switch has its own BGP AS number, and the BGP private
range is 64,512 to 65,535, which leaves you with 1023 BGP autonomous system numbers. If your IP fabric is larger than 1023 switches,
youll need to consider moving into the public BGP autonomous
system number range, which I wouldnt advise for an internal data
center, or move to 32-bit autonomous system numbers.

Chapter 6: IP Fabrics and BGP

The 32 bit takes the number up to potentially four billion ASs, with the
4,200,000,000 to 4,294,967,294 of the 32-bit range reserved for
private addressing, giving you plenty to play with. Note that while
Juniper supports 32 bit ASs as standard on its equipment, other
vendors may or do not.
Table 6.1 outlines the main points between IBGP and EBGP.
Table 6.1

iBGP and EBGP

IBGP

EBGP

Simple Design

Simple design

IBGP bound to loopbacks

EBGP bound to physical switch

Single BGP ASN

BGP ASN per a switch

Route Reflectors avoids full-mesh peering

Support for traffic engineering

Single routing domain

Distributed routing domains

Add-path required for ECMP

ECMP as standard

BGP and DC1 and DC2

Lets see how BGP ties back into our scenarios for DC1 and DC2.
In DC1 there are five rows of racks with a spine and leaf solution for
each row. If we are using EBGP per a row, the BGP design would look
like Figure 6.9.

Figure 6.9

EBGP For DC1

As you can see, each switch at both the spine and leaf layer is given its
own AS number. This can be repeated for the other rows. So, you could
have 64,512 to 523 for row one, and then increase for each device.
With DC2, as we are using an end-of-row solution, the EBGP design
would look like Figure 6.10:

Day One: Data Center Fundamentals

Figure 6.10

EBGP For DC1

Again, as outlined for the DC1 solution, you can use AS64512 and
64513 for row one and then increment by one for each device. If you
choose the IBGP method of implementation the design becomes really
easy because all you do is assign all of the components to a single AS.
This single AS could cover the whole of DC1 and you could have the
same approach for DC2 as shown in Figures 6.11 and 6.12.

Figure 6.11

IBGP For DC1

Figure 6.12

IBGP For DC2

Lets move to the support and understanding of overlay networks.

MORE? An excellent white paper on designing BGP can be found here: http://

www.juniper.net/us/en/local/pdf/whitepapers/2000565-en.pdf.

Chapter 7
Overlay Networking

Overlay and underlay correspond to the two planes in which

communication or data is sent. Fundamentally, an overlay
network is a virtual network and an underlay network is a
physical network.
An overlay network is built virtually on top of an underlay
network using VPNs, or in overlay-speak, using VTEPs (virtual
tunnel endpoints). Devices such as virtual servers, virtual switches, and others are connected to each other via these virtual or
logical links. Each link then corresponds to a path through the
virtual network and is transported over many physical or underlay networks.
As you may be aware, this decoupling of the virtual from the
physical enables faster provision of a comprehensive virtual
network for the communication between different virtual servers
without having to configure, or re-configure, the physical devices
that the virtual traffic will traverse.
The benefit for the underlay is simplicity. Because all virtual
communication is encapsulated in IP the underlay just needs to
transport simple IP data. So you can architect the underlying
physical network for simplicity, resiliency, and scale, as we have in
the previous chapters, while pushing most of the complexity to the
virtual overlay devices and servers.

Day One: Data Center Fundamentals

Tunneling Protocols
So why do we need overlay networking?
Where previously one application would be run on one physical server,
you can now carve up the processing and memory of a physical server
to be able to run multiple virtual servers, thus reducing the number of
physical servers you need while increasing the productivity of those
same servers. Your average physical server now consists of a hypervisor, or VM monitor, that can be a piece of software or firmware that
creates and runs multiple VMs.
A computer on which a hypervisor is running one or more VMs is
defined as a host machine, and each virtual machine is called a guest
machine. The hypervisor presents the guest operating systems with a
virtual operating platform and manages the execution of the guest
operating systems. Multiple instances of a variety of operating systems
can then share the virtualized hardware resources.
The applications that run on these VMs need to communicate with
each other and in general the communication is Layer 2. But why go to
the trouble of an overlay if all they need to do is talk to each other on
the same Layer 2 domain?
Overlay networking is a way to view the network as secondary to the
application. With overlay networks youre managing from the application down rather than from the network up because you want to
architect that the application gets the services and support from the
network to be able to deliver quickly and efficiently. VLANs are a
network-imposed solution with limitations, such as the number you
can have on a single switch or in a data center. A Layer 2 overlay
technology such as a Virtual Extensible LAN (VXLAN) removes those
constraints, allowing your server team to implement the network
element and bring the application to the user in a manner consistently
faster than before. From a network point of view, youre not redundant
in this process, but your concern is the Layer 3 element, which is
presented to you by the encapsulated protocol.
So, to transport traffic between the different VMs on different physical
hardware, you generally use virtual tunnels. These VTEPs typically
reside in hypervisor hosts, such as VMware Hypervisor hosts, or
kernel-based virtual machine (KVM) hosts, or they can even reside on
the switches themselves. Each VTEP has two interfaces: a switching
interface that faces the VMs in the host and provides communication
between VMs on the local LAN segment, and an IP interface that faces
the Layer 3 network. Each VTEP has a unique IP address that is used
for routing the UDP packets between VTEPs.

Chapter 7: Overlay Networking

So within each hypervisor, the VMs have the ability to communicate

with each other on the same Layer 2 segment through the use of a
virtual switch or vSwitch, but if they need to talk to another VM
located on another host, they need to traverse the Layer 3 underlay
network, which is where the VTEP is required and a protocol is used to
encapsulate Layer 2 traffic inside Layer 3 in order to pass it over the
Layer 3 underlay. As such, there are a few options.

Tunnel Encapsulation Protocols

There are a number of tunneling/encapsulation protocols that can be
used to create overlays in the data center. Network Virtualization
using Generic Routing Encapsulation (NVGRE) uses the same principle as in all of the encapsulation overlay protocols, namely it tries to
remove the limitation that comes with scaling large Layer 2 networks
by encapsulating the Layer 2 packets in a GRE tunnel. Its main
sponsor was Microsoft, but they now support VXLAN.
Another option is Stateless Transport Tunneling (STT), an encapsulation mechanism used by VMware mostly for communication between
server-based vSwitches. While STT is likely to be more efficient than
either VXLAN or NVGRE for the transfer of large amounts of information, it carries more overhead than either VXLAN or NVGRE, with
additional header information for smaller packets. Outside of VMware, its hard to find. And, VMware also supports VXLAN.
You also have the option of MPLS-over-MPLS, or MPLS-over-UDP,
but for the sake of brevity, lets assume VXLAN is our preferred
tunneling protocol as it is currently the most popular.
VXLAN encapsulates Layer 2 Ethernet frames in Layer 3 UDP packets, enabling the virtualization of Layer 2 subnets, or segments, that
can span a physical Layer 3 network.
To explain how VXLAN works, you need to understand two components:
VXLAN Network ID or VNI
And VXLAN tunnel Endpoint or VTEP.

A VNI is similar to an 802.1Q VLAN ID in that it acts as the unique

identification of that Layer 2 domain/subnet or segment and allows all
VMs within that VNI to communicate with each other.
As noted earlier, the main limitation with traditional VLANs is the
number of VLANs supported (4094), due to VLAN packet encapsulation only having a 12-bit ID space, which can only allow 4094.

Day One: Data Center Fundamentals

This limitation becomes apparent when you start to scale your data
center to support multiple customers; these customers want to have
multiple Layer 2 domains to host their applications, and they want
these applications to be active all the time, which means you may need
larger facilities to support these requirements. This is what is classed as
multitenancy.
To get around this issue, VXLAN VNI has a 24-bit length, which
means it can provide up to 16M unique L2 IDs thus supporting a
considerably larger number of L2 domain and subnets, and removing
any scaling and overlap concerns that would have been present using
VLANS.
The second element to understand is the VXLAN tunnel endpoint
(VTEP), that preforms the encapsulation and decapsulation of packets
to travel over the Layer 3 fabric.
VTEPs typically reside within the hypervisor hosts. Each VTEP has
two interfaces. One is a switching interface that faces the VMs in the
host and provides communication between VMs on the local LAN
segment. The other is an IP interface that faces the Layer 3 network.
Each VTEP has a unique IP address that is used for routing the UDP
packets between VTEPs.
A quick walkthrough might help clarify things, as shown in Figure 7.1.

Figure 7.1

A Walkthrough of VTEP

Chapter 7: Overlay Networking

As shown in Figure 7.1, VTEP1 receives an Ethernet frame from VM1,

which needs to go to VM3. Both VM1 and VM3 sit within the same
VNI, and as such, the forwarding table for that Layer 2 domain has the
MAC address for VM3, and which VTEP to pass the packet over.
VTEP1 then adds a VXLAN header that contains the VNI to the
Ethernet frame, encapsulates the frame in a Layer 3 UDP packet, and
routes the packet through to VTEP1 to VTEP 2 over the Layer 3
network.
VTEP2 decapsulates the original Ethernet frame and forwards it to
VM3. VM1 and VM3 are completely unaware of the VXLAN tunnel
and the Layer 3 network between them. As far as VM1 and VM3 are
concerned they could be sitting on the same server, or on different
servers within the same data center or in two different data centers.
To the VMs the process is transparent; they just sit on the same Layer 2
domain.
To provide a little more context, the packet diagram shown in Figure
7.2 provides some more detail about the packet structure, which will
become more relevant when we go through a larger packet walk across
a data center network.

Figure 7.2

Packet Diagram for VTEP

You can see in Figure 7.2 that the original Ethernet frame consists of
the source and destination MAC addresses, Ethernet type, and an
optional IEEE 802.1q header (VLAN ID). When this Ethernet frame is
encapsulated using VXLAN, VXLAN adds these additional elements:
A VXLAN header that comprises of an 8-byte (64-bit) field

which includes the following two fields:

Flags: 8 bits in length
VNI: A 24-bit value that provides a unique identifier for the

individual VXLAN segment.

Outer UDP Header: The source port in the outer UDP header is

dynamically assigned by the originating VTEP and the destination port is typically the well-known UDP port 4789.

Day One: Data Center Fundamentals

Outer IP Header: The outer IP header has a source IP address of

the source VTEP associated with the inner frame source (that
could be the VM or a physical server). The outer destination IP
address is the IP address of the destination VTEP that corresponds to the inner frame destination.
Outer Ethernet Header: This has a source MAC address of the

VTEP associated with the inner frame source. The destination

MAC address is the MAC address of the switching next-hop to
reach the destination VTEP.
Frame check sequence (FCS): New FCS for the outer Ethernet

frame.
In total, VXLAN encapsulation adds between 50 and 54 additional
bytes of header information to the original Ethernet frame. Because
this can result in Ethernet frames that exceed the default 1514 byte
MTU, best practice is to implement jumbo frames throughout the
network.
MORE? For a very detailed review of VXLAN, please refer to this IETF draft:

https://tools.ietf.org/html/draft-mahalingam-dutt-dcops-vxlan-00.

Packet Walkthrough Using VXLAN

Lets provide a packet walkthrough of the how a packet generated in a
VM traverses the network and the processes in place. These steps are
numbered below and are illustrated in Figure 7.3:
1. VM1 has a MAC address of 00:00:00:00:00:01 and an IP address of
10.10.10.1 and wants to send traffic to VM2.
2. This traffic is generated on VLAN 10 in the hypervisor with the
source MAC and IP and destination MAC and IP (Grey Box).
3. Host A (Hypervisor) encapsulates the original Ethernet Frame along
with the payload in to a VXLAN header (light blue) and assigns a VNI
of 1001 to act as the ID for that VXLAN. It then creates a VTEP
source IP address and destination IP address of the other end of the
VTEP, which will be on host B in the Outer IP header (orange). This is
where the control plane protocol becomes active as the learning
mechanism that allows the VTEP to discover its destination, which is
covered in the next section, Layer 2 Learning.
4. The packet is then routed across the IP fabric the same as any other
packet, as the outer packet details are same IP details as the fabric.
5. As above
6. As above

Chapter 7: Overlay Networking

7. The VTEP-B (Host-B) receives the packet and removes the VXLAN
header and forwards the original Ethernet frame.
8. The packet hits the VTEP endpoint and it strips the VXLAN header
and switches the original frame towards VM2.
9. The original frame is now just a native Layer 2 frame with payload
as was transmitted by VM1.
10. VM2 receives the original frame as if it was on the same VLAN as
VM1.

Figure 7.3

A Packet Walkthrough Using VXLAN

NOTE

Ideally, as a best practice, you start VNI numbering from 4K/5K+, for
example, VLAN 100 -> VNI 4000 or 5000. However, it doesnt
prevent you from configuring VLAN 100 -> VNI 1001, as shown in
Figure 7.3.
The steps in Figure 7.3 could easily apply to both a VTEP termination
within the hypervisor as well as a VTEP termination in a switch. Thats
because VTEP endpoints can be located within the hardware of the

Day One: Data Center Fundamentals

switch as well as within a VM, due to the simple fact that data centers
run a lot of non-virtualized or bare metal servers (BMS), which dont
have native support for VXLAN.
By placing VTEP endpoints so that they reside on the switches that BMS
servers attach to, you can allow the switch to act as a gateway between
the virtualized and non-virtualized, thus allowing traffic to move
between the two different data planes as if they were connected to the
same VLAN, as shown in Figure 7.4.

Figure 7.4

VTEP on a Switch Acting as a Gateway

Figure 7.4 shows the two VMs that reside in VLAN10 are encapsulated
to VXLAN 110, which is transported across the Layer 3 network, but
the VTEP is terminated in the switches hardware. So the VTEP is
associated with the interface that connects to the physical server, which
is configured for VLAN10, and from the point of view of both the
physical and virtual servers, they all reside on VLAN10.
Support for this feature is standard across both switches and routers
from Juniper Networks.

Layer 2 Learning
To allow VMs and BMS servers to communicate with each other there
has to be a learning mechanism in place (with associated tables) that
maps the MAC addresses of VMs and BMSs to specific VTEPs and
maintains that mapping.

Chapter 7: Overlay Networking

Lets start with the data plane first. VXLAN as a standard doesnt
include a control plane mechanism for VTEPs to share the addresses
that they have discovered in the network, but it does include a
mechanism that is very similar to the way traditional Ethernet learns
MAC addresses. Whenever a VTEP receives a VXLAN packet, it
records the IP address of the source VTEP, the MAC address of the
VM, and the VNI to its forwarding table. So when a VTEP receives an
Ethernet frame for that destination VM server on its VNI segment, it
is ready to encapsulate that packet in a VXLAN header and push it
towards that VTEP.
If a VTEP receives a packet destined for a VM with an unknown
address, it will flood and learn like a traditional Ethernet switch to see
if someone else knows the destination MAC address. But, to stop
unnecessary flooding of traffic, each VNI is assigned to a multicast
group, so the flood and learn process is limited to all of the VTEPs in
that VNIs multicast group.

A Packet Walkthrough of VTEP Flood and Learn

The next packet walkthrough will hopefully make the flooding and
learning process a little clearer.
As an example, lets suppose that VM1 in Figure 7.5 wants to send a
packet to VM2 at IP address 192.168.0.11 but does not have VM2s
MAC address. Both VMs reside in VNI 100, which has been assigned
to multicast group 239.1.1.100. The sequence follows these steps:

Figure 7.5

VM1 Sends Packets to VM2

Day One: Data Center Fundamentals

1. VM1 sends an ARP packet requesting the MAC address associated

with 192.168.0.11.
2. VTEP1 receives this ARP packet. It encapsulates this ARP request into
a multicast packet and addresses the packet to the multicast group
239.1.1.100.
3. All VTEPs in multicast group 239.1.1.100 receive the packet. They
decapsulate it and check the VNI in the VXLAN header. If the VNI for a
local VXLAN segment is 100, the VTEPs forward the original ARP
packet to that VXLAN segment. Otherwise, they drop the packet. The
VTEPs also add the mapping of the IP address of VTEP1 to the MAC
address of VM1 to their local VXLAN tables.
4. When VM2 receives the ARP packet from VTEP2, it responds with its
MAC address.
5. VTEP2 encapsulates the response in a unicast IP packet and sends it to
VTEP1.
6. VTEP1 receives the ARP packet, decapsulates it, and passes it on to
VM1. VTEP1 now stores the mapping of the VTEP2 IP address and the
VM2 MAC address in its VXLAN mapping table.
At this point, all relevant MAC addresses have been learned, and VM1
and VM2 can communicate directly via unicast in the future.
Typically, to reduce unnecessary flooding of packets, administrators
assign each VNI to its own multicast group. However, there is no requirement that each VNI have its own multicast group you can assign
multiple VNIs to a single multicast group. Isolation of the VXLAN
segments is maintained in this case because the VTEP always checks the
VNI before it forwards decapsulated packets to a VXLAN segment.
Before moving on to the control plane, there is one aspect that is worth
covering: Layer 3 routing for VXLAN because the subject comes up quite
often due to differences in support for this feature within different
platforms.

VXLAN Routing
At its most basic, VXLAN routing is the ability for one VLAN to talk to
another VLAN. To allow this to happen, you need a Layer 3 gateway. Its
the same in an overlay network, but when a VXLAN VNI needs to talk
to another VXLAN VNI, you need to route between them and you need
a Layer 3 gateway.
Juniper prefers to implement VXLAN routing natively in hardware as
opposed to either using a multi-stage solution or cabling, either internally

Chapter 7: Overlay Networking

or externally, that recirculates the packets in and out of a switch in

order to achieve the same process. If you take the spine and leaf model
from previous chapters, Layer 3 VXLAN routing can be done at both
the spine and leaf layers. In the past, this hasnt been possible at the leaf
layer due to limitations with merchant silicon, but those limitations
have now been removed, thus allowing routing on a top-of-rack that
sits at an affordable price point. Spine layer VXLAN routing has been
supported for awhile now in the QFX10000 Series, the MX Series, or
the EX9200 Series all of which use Juniper silicon, which doesnt have
those limitations.
The process for routing VXLAN is the same at either the leaf or the
spine. Lets quickly walk a packet through the process, beginning with
Figure 7.6.
Using the exploded diagram of a single switch in Figure 7.6, there is a
server attached to this QFX switch; this server is in VLAN 100 with an
IP address of 10.1.100.10 and a gateway of 10.1.100.1. It needs to send
a packet to another server on VLAN 101 with an IP address of
10.1.101.10.

Figure 7.6

Packet Walkthrough of VXLAN Routing

Configured on the switch is a VTEP with the VLAN of 100, a VNI

of 1000, and a Layer 3 IRB (integrated routing and bridging) of 100.
On the same switch there is also another VTEP with a VLAN of 101,
a VNI of 1001, and its own IRB of 101, which ties in to our destination
server. Finally there is a secondary Layer 3 virtual routing and forwarding (VRF) that the two VNIs connect and where the routing is done.
1. The QFX switch receives the VXLAN packet with the outer
destination IP of its own VTEP. It decapsulates the VXLAN, which
leaves a native packet.

Day One: Data Center Fundamentals

2. The QFX switch does a lookup on the destination MAC, which is

the local IRB VIP MAC.
3. Another Layer 3 lookup is done inside the L3 VRF routing table.
4. An ARP lookup for 10.1.101.10 (the IP address of our destination
packet). If it already exists then destination MAC is resolved. If ARP
does not exist it will be resolved by looking at the MAC table of the
other VNI.
5. The QFX10000 switch then generates a new Layer 2 header with
the destination MAC for 10.1.101.10 (the destination server) and then
forwards the encapsulated VXLAN packet to the remote VTEP.
That, in a nutshell, is how Juniper routes between VXLANs on a single
switch in the same PFE/ASIC. Its good to know. Now lets move on to
the control plane.
IMPORTANT When you need detailed or up-to-date information on any aspect of

this fundamentals book, start at Junipers Tech Library where all the
products, protocols, and configurations can be found: http://www.juniper.net/documentation. The data sheets for all the Juniper devices
mentioned in this Day One book can be found on their specific
product pages at http://www.juniper.net/us/en/products-services/.

Applying What You Have Learned to Your Design

So far weve covered the breakdown of VXLAN and how a packet
traverses the network, but how does this apply to the DC1 and DC2
scenario designs? Starting with DC1 if you were to select QFX5100
Series switches you can support all of the elements outlined here with
the exception of VXLAN routing. For VXLAN routing, you would
need to pass those packets to the spine layer, which would no doubt be
a QFX10000 Series. The other option is to implement the newer
QFX5110 series, which does support native VXLAN routing.
In DC2, because a QFX10000 Series chassis was specified as the
end-of-row devices, VXLAN routing is supported as is VXLAN VTEP
termination in hardware. In both cases this means that both of the data
center designs support overlay networking from day one.
This leads to the next method of Layer 2 learning via the control plane,
which is divided into two elements: a controller-based Layer 2 control
plane and going controller-less.
Chapter 8 covers controller-based solutions including Juniper Contrail
and VMwares NSX, while Chapter 9 covers EVPN, which the author
would argue is the perfect controller-less based protocol for data
centers.

Chapter 8
Controllers

The overlay in data centers can either be constructed by a controller or independently by the network nodes. In a controller-based
solution, a central brain such as Juniper Contrail or VMwares
NSX solution holds the Layer 2 tables and knows how to reach all
of the elements in the virtual network, as shown in Figure 8.1. In a
controller-less based solution, you are reliant on a protocol to
distribute from one point to another.

Figure 8.1

A Controller Run Data Center

When a VM needs to send traffic to another VM somewhere else

in the data center that it doesnt know how to get to, it does a
lookup via a vSwitch or vRouter that resides in the hypervisor of
that server. The vRouter (same as a vSwitch) then performs a
lookup on its central controller and receives the routes that allow

Day One: Data Center Fundamentals

a path to be constructed across the IP fabric allowing those two VMs

to talk to each other. The benefit of this solution is that its very
dynamic and removes the need to configure any network elements, as
they just need to route traffic at the IP layer.
It also means that the same controller can configure the switches in the
path via protocols such as Open vSwitch Database (OVSDB), allow
the MAC routes learned in the physical layer to be passed to the
controller, and allow the controller to distribute learned forwarding
routes into the physical layer.
So how does a controller like Contrail work to support its traffic
solutions?
First, lets quickly review how a controller-less network uses an
underlying control plane protocol, such as EVPN, to provide Layer 2
MAC reachability, because EVPN is more likely to be configured on
the physical network hardware as opposed to the controller-based
solution, which sits at the overlay layer. Examine Figure 8.2.

Figure 8.2

A Controller-Less Data Center

As you can see in Figure 8.2, the controller-less network provisions the
VTEP network from the network switches. VM-based servers would
connect to these switches via standard Layer 2 trunks. While these
VTEPs can be controlled via a management platform such as Junos
Space, a controller-less solution is a little more static in its implementation and management.
EVPN is covered in detail in Chapter 9; its a complex subject and
requires a diverse protocol in its support for different architectures.
Now, lets get back to a controller-based solution. How does it work?
Lets look at Juniper Contrail as an example.

Chapter 8: Controllers

Junipers Controller-Based Solution: Contrail

Contrail is a simple, open controller-based solution that automates and
orchestrates the overlay network. It consists of two main components,
the Controller and the vRouter, and takes normal networking terms
such as MPLSoGRE, MPLSoUDP, and VXLAN with BGP and applies
them to the virtual world.
The Contrail controller is a logically centralized but physically distributed (distributable) software-defined networking (SDN) controller,
responsible for providing the management, control, and analytics
functions for the overlay network.
The Contrail vRouter acts as the local forwarding plane running in the
hypervisor of a virtualized server. It takes the networking principles
from physical routers and switches in a data center and puts them into
a virtual overlay network hosted in the virtualized network. (The
vRouter is very similar to a vSwitch but it also provides additional
higher layer services such as routing and security, hence the vRouter
name.)
The Controller provides the centralized control plane and management
plane of the system and orchestrates the vRouters, which forward the
traffic. The physical underlay network is responsible for providing
unicast IP connectivity from any physical device (server, storage device,
router, or switch) to any other physical device. The vRouters running
in the hypervisors of the virtualized servers create a virtual overlay
network on top of this physical underlay network using a mesh of
dynamic tunnels among themselves. In the case of Contrail, these
overlay tunnels can be MPLS over GRE/UDP tunnels, or VXLAN
tunnels, which means support for both Layer 3 overlays using the same
protocols as standard MPLS and Layer 2 VXLAN with EVPN.
The underlay physical routers and switches do not contain any state
relating to MAC addresses, IP address, or policies for VMs. The
forwarding tables of the underlay physical routers and switches only
contain the IP prefixes or MAC addresses of the physical servers.
The vRouters, on the other hand, contain a separate forwarding table
for each virtual network. That forwarding table contains the IP
prefixes to support Layer 3 overlay networks or the MAC addresses
for Layer 2 overlays of the VMs. Because of the central controller, no
single vRouter needs to contain all of the addressing for all the VMs in
the entire data center. Figure 8.3 illustrates a controller-based solution
using Juniper Contrail.

Day One: Data Center Fundamentals

Figure 8.3

A Controller-Based Solution Using Juniper Contrail

Before examining the vRouter and controller in more depth, lets

quickly cover three aspects of the Contrail system that make the solution work: configuration, control, and analytics, all of which are present
in both the vRouters and controller.
Configuration: The configuration component keeps an active copy

of the configuration state, present in each vRouter and in the

controller.
Control: The control component implements a logically-central-

ized control plane that is responsible for maintaining network

state. This control function interacts with each vRouter and with
other network elements, such as gateways, to ensure that network
state is consistent.
Analytics: The analytics component collects, stores, correlates,

and analyzes information from network elements, virtual or

physical. This information includes statistics, logs, events, and
errors. This information aids in troubleshooting and real-time
event data for capacity planning.

Chapter 8: Controllers

Contrail Controller
From the Controllers point of view, each of the three elements:
configuration, control, and analytics, becomes its own node. So theres
a control node, a configuration node, and an analytics node. Each of
these logical nodes runs on an X86 processor that may be on a separate
physical server or running as a VM, or indeed, all running on a single
server in your lab. As all nodes run in an active/active configuration,
you can run multiple instances of these nodes so no one node becomes
a bottleneck, and that allows you to scale out the solution and provide
redundancy. These nodes are interconnected to each other, with the
control nodes being the nodes that the vRouters interact with as shown
in Figure 8.4.

Figure 8.4

Contrail Controller

Each vRouter connects to three or more controller nodes (as shown in

Figure 8.4) where all the nodes are active-active. Think of this as a
collection of control nodes where the vRouter receives all its state
(routes, routing instance configuration, etc.) from each of these control

Day One: Data Center Fundamentals

nodes. The vRouter then makes a local decision about which copy of
the control state to use its similar to how a BGP PE router receives
multiple copies of the same route (one from each BGP neighbor) and
makes a local best route selection. The information is then added to the
local table until the controller advertises a better route.
If a control node fails, the vRouter agent will notice that the connection to that control node is lost. The vRouter agent will flush all routes
that it received from the failed control node. It already has a redundant
copy of all the states from the other control node. The vRouter can
locally and immediately switch over without any need for resynchronization. The vRouter agent will do a discovery for the remaining
control nodes to see if a new control node exists to replace the old one.
If not it will continue to use the single control node that it knows about
until another one recovers.

Contrail vRouter
The vRouter sits within the compute servers hosting your VMs. The
standard configuration assumes Linux is the host OS, and KVM is the
hypervisor (but other hypervisors are supported). The vRouter is then
divided into two functions, the vRouter forwarding plane and the
vRouter agent. The vRouter forwarding plane for Virtio sits in the
Linux kernel, but for those running the Data Plane Development Kit
(DPDK) support it sits in user space and the vRouter agent is in the
local user space as shown in Figure 8.5.

Figure 8.5

Structure of the Contrail vRouter

Chapter 8: Controllers

The vRouter agent connects to the control nodes to receive forwarding

information and the vRouter forwarding plane is the virtual router.
The vRouter agent is a user space process running inside Linux. It acts
as the local, lightweight control plane, and is responsible for the
following functions:
Exchanging control state such as routes with the Control nodes
Reporting analytics state such as logs, statistics, and events to the

analytics nodes
Installing routing tables into the forwarding plane
Discovery of existing VMs and their associated attributes
Applying policy for the first packet of each new flow and install-

ing a flow entry in the flow table of the forwarding plane

Proxy DHCP, ARP, DNS, and MDNS

Each vRouter agent is connected to at exactly two control nodes for

redundancy in an active-active redundancy model.
As mentioned, the vRouter is a virtual router and runs as a kernel
module or in user space depending on the forwarding mechanism
(Virtio or DPDk) in Linux and is responsible for the following functions:
Encapsulating packets and de-encapsulating packets sent and

received over the overlay network

Assigning packets to a routing instance
Assigning packets received from the overlay network and

assigning them to a routing instance based on the MPLS label or

virtual network identifier (VNI)
Doing a lookup of the destination address in the forwarding

information base (FIB) and forwarding the packet to the correct

destination. The routes may be Layer 3 IP prefixes or Layer 2
MAC addresses. Optionally, when applying forwarding policy
using a flow table it matches packets against the flow table and
applies the flow actions.
One element of this process thats important to discuss is how you can
assign different VMs to different vRouter instances as shown in Figure
8.6. This is relevant in the same way that a VLAN can be assigned to a
VXLAN tunnel, and as such, a Layer 2 domain. The principle is the
same, but you create separate virtual routers to do the isolation and

Day One: Data Center Fundamentals

move it down in to the hypervisor layer, as opposed to being on the

switch layer. That means you can have multiple Layer 2 and Layer 3
domains, which will not interact with each other unless configured to
do so.

Figure 8.6

Assigning Different VMs to Different vRouters

Where Figure 8.6s detail becomes naturally advantageous is in a multitenant environment, where you need to offer multiple VMs and
domains but with complete separation and simple centralized control
and configuration.
As mentioned previously, Contrail vRouters, and Contrail as a whole,
support MPLS over GRE or UDP and VXLAN, but it is worth noting
that the selection of which data plane protocol to use is based on a
preference order that can be defined during setup, and also takes into
account the capabilities of the two endpoints of the tunnel. Either you
define the tunnel type or the vRouters can do that based on what the
tunnel endpoints support.
So elements covered in Chapter 7 on VXLAN stay the same and you
implement Contrail to manage that overlay element. If, on the other
hand, you come from an MPLS background, then the support for
MPLS is the same as it would be for a MPLS WAN network, but
implemented in the data center where the encapsulation method of the
packets is MPLS with a MPLS label to act as its ID.
Lets join this all together with a packet walkthrough on how VM1A
could talk to VM1B across our fabric.

Chapter 8: Controllers

Packet Walkthrough in a Contrail Solution

In the scenario shown in Figure 8.7, VM1A needs to send packets to
VM1B, which is located on a different server in the data center.

Figure 8.7

Packet Walkthrough for Contrail vRouter Traffic

Lets begin with an application sitting within VM1 that wants to

communicate with another application within VM2:
1. VM1 has a default route pointing to a local address that resides in
the vRouting instance 1, effectively its next-hop address, so it sends an
ARP request to that next-hop address.
2. The Routing 1 instance receives the ARP, and the proxy process
within the vRouter 1 instance replies.
3. VM1 receives the reply and sends the initial packet through to
vRouter 1.
4.The vRouter 1 instance receives the packet and does an initial lookup
in its local FIB table. If the route is in the local FIB table, then the
packet would be encapsulated and sent on to the next hop, which
would be the global FIB table that resides in vRouter 1 (vRouter 1 can

Day One: Data Center Fundamentals

have multiple routing instances). The instances look after their

associated VMs and provide local routing and switching between VMs in
that instance. If the packet is destined for outside of that instance, if the
packet is destined for a node outside the routing instances domain, then
one of three things will happen:
If a policy exists that says that the destination should be reach-

able from that routing instance, then a route will be applied in that
instance providing the next-hop and label required to reach the
destination and normal encap as outlined;
If theres a default route, the packet will follow that path;
If nothing is defined, then the packet will be dropped.

5. The vRouter 1 does a lookup of the new destination IP address of the

encapsulated packet in its global IP FIB table. If its present then it will be
encapsulated again with the outer IP address of the compute server
where the hypervisor resides. It does a lookup in the master global route
table, responds back to vRouter 1 with the routes, and vRouter1 updates
its tables.
6. The encapsulated packet is now transported over the IP fabric until it
gets to compute server 2. It decapsulates the packet and passes it to the
vRouter, which does a lookup in its global FIB. It sees that the packet is
destined for vRouter instance 1 and passes it to that instance.
7. Finally, vRouter instance 1 receives the packet, decapsulates the
packet to expose the inner IP or Layer 2 information, then checks its
local FIB to route or switch that packet to the correct VM, and sends it
through to that VM.
So, thats Contrail. Youll see similarities with a router that supports multiple VRFs that access a central control plane holding the routing and
switching information. It is also similar to BGP routing devices that are
clients of a BGP route reflector holding all the routes for that BGP AS.
MORE? For an even more detailed and up-to-date view of Juniper Contrail and

its specifications, please refer to its product page on the Juniper web site,
with links to papers, documentation, and solutions: http://www.juniper.
net/us/en/products-services/sdn/contrail/.
MORE? If you require more detail on how VMWare NSX and Juniper work

together to provide a more enterprise oriented solution, refer to the NSX

design guide: https://www.vmware.com/files/pdf/products/nsx/vmw-nsxnetwork-virtualization-design-guide.pdf. And the Juniper version can be
found here: https://www.juniper.net/us/en/local/pdf/
whitepapers/2000525-en.pdf.

Chapter 8: Controllers

Contrail as a Controller in a Pure Switch Solution

This chapter has been looking at SDN from the point of view of
encapsulation from the hypervisor layer. Contrail also supports a
dynamic overlay and encapsulation based purely in the switch layer.
This is to provide support for BMSs that dont run VM-based services,
as well as for virtual servers for which you may not want the encapsulation to happen at that layer, but still want a dynamic central controller for MAC learning and provision.
Contrail achieves this through the use of the Open vSwitch Database
Management (OVSDB) Protocol that is supported in all of the QFX
switches. The OVSDB protocol is used to configure the top-of-rack
switch and to import dynamically-learned addresses from the BMS and
virtual servers connected to those switches. Contrail then uses VXLAN
encapsulation as the data plane communication across the fabric
between the top-of-rack switches.
To implement this overlay, Contrail employs two services: a top-ofrack services node (TSN) and a top-of-rack Agent. TSN acts as the
multicast controller for the top-of-rack switches providing the MAC
learning and distribution. The TSN also provides DHCP and DNS
services to the BMS or virtual instances running behind top-of-rack
switch ports.
The TSN receives all the broadcast packets from the top-of-rack
switch, and replicates them to the other switches within the fabric and
the required compute nodes in the server cluster. Broadcast packets
from the VMs in the server cluster are sent directly from the respective
compute nodes to the top-of-rack switch. This allows for a full map of
the server and switch topology to be constructed.
As mentioned, TSN can also act as the DHCP server for the BMS or
virtual instances, leasing IP addresses to them, along with other DHCP
options configured in the system. The TSN also provides a DNS service
for the BMS. Multiple TSN nodes can be configured in the system
based on the scaling needs of the cluster.
The other service, a top-of-rack agent, is provisioned in the Contrail
controller cluster and acts as the OVSDB client for the top-of-rack
switch, and all of the OVSDB interactions with the top-of-rack switch
are performed by using the top-of-rack agent. The top-of-rack agent
programs the different OVSDB tables onto the top-of-rack switch and
receives the local unicast table entries from the top-of-rack switch.
The typical practice is to run the top-of-rack agent on the TSN node.
In addition to running the different tables, the top-of-rack agent also
receives the configuration information for the top-of-rack switch from

100

Day One: Data Center Fundamentals

Contrail. The top-of-rack agent translates the Contrail configuration

to OVSDB and populates the relevant OVSDB table entries in the
top-of-rack switch.
Control Plane

At the control plane level we can use EVPN to allow us to contrast

multiple virtual networks (if required for, say, a multi-tenant solution).
The top-of-rack agent receives the EVPN route entries for the virtual
networks in which the top-of-rack switch ports are members, and adds
the entries to the unicast remote table in the OVSDB.
MAC addresses learned in the top-of-rack switch for different logical
switches (entries from the local table in OVSDB) are propagated to the
top-of-rack agent. The top-of-rack agent exports the addresses to the
control node in the corresponding EVPN tables, which are further
distributed to other controllers and subsequently to compute nodes
and other EVPN nodes in the cluster such as gateways for transportation between data centers.
The TSN node receives the replication tree for each virtual network
from the control node. It adds the required top-of-rack addresses to the
received replication tree, forming its complete replication tree. The
other compute nodes receive the replication tree from the control node,
whose tree includes the TSN node, thus allowing every device in the
fabric a complete view.
Data Plane

In the data plane you can use VXLAN encapsulation (as discussed in
pervious chapters). The VTEP for the BMS is on the top-of-rack
switch. Unicast traffic from BMS is VXLAN-encapsulated by the
top-of-rack switch and forwarded, if the destination MAC address is
known within the virtual switch. Unicast traffic from the virtual
instances in the Contrail cluster is forwarded to the top-of-rack switch,
where VXLAN is terminated and the packet is forwarded to the BMS.
Broadcast traffic from BMS is received by the TSN node which uses the
replication tree to flood the broadcast packets across that specific
virtual network. Broadcast traffic from the virtual instances in the
Contrail cluster is sent to the TSN node, which then replicates the
packets to the top-of-rack switches.

Chapter 8: Controllers

DC1 and DC2 Contrail Solutions

Remember the solutions we crafted earlier for DC1 and DC2? Well,
both DC1 and DC2 topologies support a controller-based solution
using either Contrail or VMware, whereby the VTEP termination can
be done either in the hypervisors of servers (KVM or ESXi based), or
directly in the switch hardware for BMS-based and HyperV systems. In
either case, the topology supports the IP fabric and interaction between
the controller-based systems.
Figure 8.8 shows DC1 with a controller acting as the control plane for
the Layer 2 virtual networks between the servers.

Figure 8.8

DC1 Contrail Solution

The controllers points of control (represented with red dashed lines)

attach to the servers where software VTEPs are constructed for the
multiple VMs as well as in the switch hardware, where the controller
can construct VTEPs and where the spine switches act as gateways
between the virtual overlay and wider connectivity all the way up to
the core layer (the controller, in both instances, could use the core layer
as the gateway if required; its an option).
Remember that the principles are the same in the DC2 topology; the
top-of-rack termination was just moved from the racks to two end-ofrow chassis switches. The VTEP termination in hardware is still fully
supported as is the software VTEP pass-through to the hypervisor
based servers.

101

102

Day One: Data Center Fundamentals

Figure 8.9

DC2 Contrail Solution

You can see in DC2 that traffic is more centralized due to the end-ofrow design. IIt also means that the gateway between the wider network
and the overlay sits at the spine/core acting as the main aggregation
point between all of the rows, and then onward connectivity to the
WAN or outside of the data center.
Okay, remember all this because next up is EVPN as a controller-less
based solution for supporting overlay networking.

Chapter 9
EVPN Protocol

As mentioned in Chapter 8, the second option for control planes

in an overlay network is to remove the controller and use a
protocol to provide the MAC learning mechanism for your
VXLAN VTEPs.
The choice is somewhat limited to multicast or EVPN. While
multicast is a valid option, it is a limited protocol compared to
EVPN and the additional topologies EVPN can support. So lets
stick with EVPN-VXLAN.
MORE? Note that EVPN addresses lots of different implementations

besides Level 2 control plane as outlined in this chapter. For more

details on all the different implementations of EVPN look in the
Juniper TechLibrary: http://www.juniper.net/techpubs/en_US/
junos/topics/concept/evpns-overview.html.
EVPN addresses two issues: the first, as we have discussed, is a
MAC learning control plane for overlay networks, and the second
is the need for workload mobility. Remember that workloads, or
applications, require a Layer 2 domain to interact with each other.
Thats fine inside a single data center where you can stretch
VLANs either in a traditional sense or via an overlay. But in many
cases those same workloads need to be present in two (or more)
data centers to provide an active/active redundancy to client
applications. That means stretching these VLANs over a WAN
between the two data centers and making that Layer 2 domain
seem as if it is locally present. But thats really the focus of EVPNVXLAN stitching to EVPN-MPLS. Lets focus on EVPN-VXLAN.

104

Day One: Data Center Fundamentals

For the past few years you could use VPLS (virtual private LAN
service) to stretch that Layer 2 domain between sites. But while VPLS
did a good job, like any protocol it also came with limitations regarding: MAC address scaling, support for multicast in a sensible way,
multi-homing active/active, transparent customer MAC address
transport, faster convergence, and no doubt the largest pain, ease of
management.
EVPN attempts to address these issues, but remember that its still a
new protocol and in some cases the standards are still being worked
out, which is why youll see slightly different implementations by
different vendors.
EVPN is in the BGP family. It uses multi-protocol BGP for the learning
of MAC addresses between switches and routers and allows those
MAC addresses to be treated as routes in the BGP table. This means
you can use multiple active paths both inside and between data centers
without blocking links. But youre not just limited to MAC addresses.
You can you use IP addresses plus the MAC address (this forms a ARP
entry) to be routed and you can combine them further with a VLAN
tag as well.
Given this flexibility for both Layer 2 and Layer 3 addressing, and the
fact that you can use a single control plane such as BGP for both the
internal network and the external WAN, the benefits of EVPN quickly
become apparent.
Before delving into the innards of EVPN, lets sync up and run through
some of the terminology and how that terminology applies to a fabric.
Lets start with Figure 9.1, which shows two servers with two VMs per
server attached to leaves in a standard spine and leaf topology. Compared to previous diagrams, you should note that that each server is
attached to two leafs for resiliency, just as it should be in the real
world. The diagram used in Figure 9.1 is used throughout this chapter
to build upon EVPN concepts and where they apply.
Starting with the server connection, as shown in Figure 9.2, the server
connects to the switch and has a trunk with two VLANs present
(VLAN 1 and VLAN 2). In EVPN-speak these two VLANs are classed
as Ethernet Tags, which is simply the identity of the VLAN, so Ethernet Tag 1 corresponds with VLAN Tag 1 and in EVPN the Ethernet tag
maps to the VLAN tag.

Chapter 9: EVPN Protocol

Figure 9.1

Basic EVPN Topology

Figure 9.2

EVPN Tags and Mapping

105

106

Day One: Data Center Fundamentals

Inside each EVI is MAC-VRF, which is a virtual MAC table for the
forwarding and receiving of MAC addresses within the EVI. You also
have an import policy and an export policy.
The VRF export policy for EVPN is a statement configured in the VRF.
This statement causes all locally learned MACs to be copied into the
VRF table as EVPN Type 2 routes. Each of the Type 2 routes associated
with locally learned MACs will be tagged with the community target of
say 1:1, and these tagged routes are then advertised to all switches in the
fabric.
The VRF import policy statement does the reverse of the export statement to accept routes that are tagged with that target community.
You also have a route distinguisher or RD that is assigned to the MACVRF, again this is unique, and its ID is advertised into the BGP control
plane that runs across the whole of our fabric.
There is a route target (RT) community. Each EVPN route advertised by
a switch in the fabric contains one or more route target communities.
These communities are added using VRF export policy or by a configuration, as mentioned earlier. When another switch in the fabric receives
a route advertisement from other switches, it determines whether the
route target matches one of its local VRF tables. Matching route targets
cause the switch to install the route into the VRF table whose configuration matches the route target.
Finally, EVPN gives you the flexibility to support different VLAN
mapping options per an EVP instance. These different supported
options are classed as EVPN services.

VLAN Services
There are three VLAN services with the first one being VLAN-based
service as shown in Figure 9.3. With this service a VLAN is mapped to a
single EVI and it becomes the EVI for that VLAN across the fabric.

Figure 9.3

VLAN-Based Service

Chapter 9: EVPN Protocol

This method provides excellent separation on a VLAN-per-VLAN

basis and limits the MAC broadcast for each VLAN, but scale and
operational overhead are the limiting factors, especially if you want
several thousand VLANs per data center and their associated EVIs.
One benefit that is worth considering is something called VLAN
manipulation or normalization. This means that you can map the
original VLAN to a different VLAN, something you can do with VPLS.
An example would be if you had two fabrics in a single data center,
with odd numbered VLANs in one fabric and even numbered VLANs
in the other. You can map VLAN 1 to EVI-1 and then to, say, VLAN
11 in the other fabric.
Your next VLAN service option, which was outlined in the initial
example, is mapping two or more VLANs to a single EVI. This is called
VLAN bundle service. It means that a group of servers with a series of
VLANs can be mapped to a single EVI across the fabric as shown in
Figure 9.4. This is useful if you are offering tenant-based services and
tenants have overlapping VLANs with other tenants. Providing
VLANs with a single EVI means they have a unique RT and RD across
the fabric providing complete separation.

Figure 9.4

VLAN Bundle Service

The benefits here are the efficient way in which you can bundle like
VLANs together and operationally make the configuration a lot easier.
But traffic flooding will affect every VLAN in the EVI, and you have no
support for the VLAN mapping, so you are sharing the EVI with lots
of other VLANs.
The last service is called VLAN aware service and it allows for multiple
VLANs and bridge domains to be mapped to a single EVI, as shown in
Figure 9.5. It allows all of the VLANs to share the single EVI but
because you have the one-to-one mapping of VLAN to bridge domain,
you can assign an ID or label to that bridge domain to provide separation between other VLANs. It also means that flooding only affects the
VLAN in which it occurs, as opposed to the bundle service where it
affects every VLAN in the EVI.

107

108

Day One: Data Center Fundamentals

Figure 9.5

VLAN Aware Service

NOTE

At the time of this books writing, the Junipers QFX Series currently
only supports VLAN Aware and VLAN Bundle. The MX Series and
the EX9000 Series support all three.
While you have already read about data plane encapsulation like
VXLAN and the different protocols that you can use for the overlay,
its still worth touching upon the ones supported by EVPN.

Data Plane
As shown in Figure 9.6, there are currently five supported data plane
technologies for EVPN. They are MPLS, PBB, SST, NVGRE, and
VXLAN. All five are encapsulation methods using EVPN and BGP as
the common control plane.

Figure 9.6

EVPN Supported Data Plane Technologies

At the time of publication of this Day One book, Juniper supports

three forms of EVPN encapsulation: VXLAN, MPLS over MPLS, and
MPLS over UDP.
You know that the VXLAN encapsulation method over the fabric uses
the concept of VNIs or VXLAN Network IDs, so your VLAN 1 and 2
are assigned VNI-1 and VNI2.

Chapter 9: EVPN Protocol

You should now have two VLANs that have assigned two VXLAN
VNIs with their MAC addresses learned and advertised in EVPN EVI
1, and assigned a RD of 1, that then provides its unique ID across the
fabric.
Okay, but one of the elements that you have to consider is EVPN
routing in BGP. It relates to the various types of reachability information that EVPN will advertise into BGP. In the wonderfully named
IETF document: BGP MPLS Based Ethernet VPN (https://tools.ietf.
org/html/draft-ietf-l2vpn-evpn-11), this information is classed as
Network Layer Reachability Information (NLRI).

BGP Route Types for EVPN

The need for NLRIs are to allow two BGP speakers in the fabric, which
in this case are switches, to advertise their capabilities and make sure
theyre compatible so they can both communicate with and transport
traffic between each other.
At the time of publication of this book, there are five route types
supported for EVPN by Juniper Networks:
1. Ethernet Auto-Discovery (AD) Route: used for multi-path and mass
withdraw.
2. MAC/IP Advertisement Route: used for MAC advertisements.
3. Inclusive Multicast Ethernet Tag Route: used for BUM flooding
(Broadcast, Unicast, and Multicast).
4. Ethernet Segment Route: used for ES (Ethernet Segment Discovery)
and DF (Distributed Forwarder) Election.
5. IP Prefix Route: used for IP route advertisement.
Lets run through each route type in a little more detail in order to
understand how EVPN interacts with BGP and provides transport over
the fabric.

EVPN Type 1 Route: Ethernet Auto-Discovery (AD) Route

Auto-discovery is the process by which a leaf switch with a new route
advertises the new route to the fabric and to all the other leaf switches
that are part of the same EVPN segment. The advertised route has an
ESI value, as well as an extended community value, so other leaves
know which EVPN segment it belonged to. The community contains a
flag that lets other switches in the fabric know if the traffic can be
load-shared over multiple links or if its a single link. If the flag is set to
1, that means only one link associated with the Ethernet segment can

109

110

Day One: Data Center Fundamentals

be used for forwarding. If the flag is set to 0, that means that all links
associated with the Ethernet segment can be used for forwarding data.
Basically its the difference between active/active links or single active.
So if a remote leaf, (say Leaf 4, in Figure 9.7) receives an Ethernet
auto-discovery route from Leaf 1 and Leaf 2 (as the server is dualattached), it will look at the advertisement, see that its for the RED
EVPN segment (which it is a part of) and install that route into its
table. It also knows it will have two VXLAN tunnels to forward traffic
back over to Server 1, as both leaves and the VXLAN tunnels are in the
same EVI (RED).

Figure 9.7

EVPN Type 1 Route

One benefit of the auto-discovery process is that in the event of a link

failure convergence times can be a lot faster. Usually, when a link fails
to a server, a leaf switch would withdraw each of its individual MAC
Advertisements from that server. If that switch is holding thousands of
MAC addresses for that link, the normal withdrawal process would
mean issuing thousands of withdrawal notifications for all of those
MAC addresses.
Because an auto-discovery route is associated with an interface as
opposed to a single MAC address, a single withdraw route statement,
say from Leaf 1 to Leaf 4, would tell Leaf 4 to remove all of the those
MAC addresses it learned from Leaf 1. This means your convergence
times are significantly reduced.

Chapter 9: EVPN Protocol

EVPN Type 2 Route: MAC/IP Advertisement Route

The purpose of a Type 2 route is to advertise MAC addresses it but it
can also advertise IP addresses that are bound to the same MAC
address. Take a look at Figure 9.8.

Figure 9.8

EVPN Type 2 Route

Using Figure 9.8, lets say that Leaf 4 learns the MAC addresses in the
data plane from Ethernet frames received from Server 2. Once Leaf 4
learns Server 2s MAC address, it automatically advertises the address
to other leaves in the fabric and attaches a route target community,
which is the solid red circle in Figure 9.8.
Upon receiving the route, Leaf 1 has to make a decision as to whether
it should keep the route. It makes its decision based on whether an
import policy has been configured to accept red route targets. No
policy, then the advertisement would be discarded.
So, at a minimum, each EVI on any participating switches for a given
EVPN must be configured with an export policy that attaches a unique
target community to the MAC advertisements, and also, it must be
configured with an import policy that matches and accepts advertisements based on that unique target community.

111

112

Day One: Data Center Fundamentals

EVPN Type 3 Route: Inclusive Multicast Ethernet Tag Route

To understand EVPN Type 3 routing in EVPN, you need to understand BUM traffic. BUM traffic is really broadcast traffic that in a
normal network would be flooded to every node in the same VLAN/
broadcast domain. In a Juniper Networks solution using EVPN/
VXLAN, the switches in the fabric support ingress replication of
BUM traffic, which means when BUM traffic arrives on a switch, that
switch unicasts copies of the received BUM packets to each switch
that belongs to the same EVPN instance, as opposed to just broadcasting the traffic everywhere.
Type 3 routes inform the remote switches in the fabric how BUM traffic should be handled. This information is carried in an attribute or
feature within the VXLAN tunnel called Provider Multicast Service
Interface (PMSI). This attribute specifies whether PIM (Protocol
Independent Multicast) or ingress replication should be used to send
BUM traffic.
This EVPN route is very simple. The route informs remote PEs of
how BUM traffic should be handled and the information is carried in
the PMSI Tunnel attribute. It specifies whether PIM or ingress
replication will be used and the addressing that should be used to
send the BUM traffic. So Leaf 2 advertises that it is expecting and
using ingress replication and that Leaf 1 should use 4.4.4.4 as the
destination address of the VXLAN packets that are carrying BUM
traffic.

EVPN Type 4 Route: Ethernet Segment Route

Type 4 Routing solves two complications with overlay networking
first it helps in the designated forwarder election process and second,
it helps add a new split horizon rule.
Okay, but what is designated forwarding and what is split horizon?
Split horizon is the method of preventing routing loops by routing
protocols from advertising a route back onto the interface from
which it was learned.
Standard EVPN already has some default split horizon rules in place.
For example, if a switch receives a BUM packet from a local server:
It floods to servers in the same VLAN
Floods to remote switches in the same VLAN
But it will not flood to the original server that sent the BUM

packet

Chapter 9: EVPN Protocol

If a switch in the fabric receives a BUM packet from another switch in

the fabric:
It floods to local servers in the same VLAN
But wont flood to other switches in the fabric

The problem is that active/active connections across our fabric can

break these rules. Earlier, in Type 1 Routing Ethernet Auto-Discovery,
you learned that Type 1 can enable multipath forwarding when a
server is dual-connected to two switches and when you are sending
traffic over multiple links in a fabric. But while this is fine for unicast
traffic, its a little more problematic with BUM traffic.
For example, if Leaf 1 in Figure 9.8 makes a copy of the received BUM
traffic from Server 1 and then sends a unicast copy of this packet to all
leaf switches in the same EVPN instance, there is a likelihood that
Server 2 will receive multiple copies of the same packet. The other
potential issue is that you could start creating loops as traffic is forwarded back to the source. Again, using Leaf 1 as an example, it
receives a BUM packet and sends a unicast copy to all switches in the
same EVPN instance. Leaf 2 receives this packet and because of the
split horizon rule forwards it back to the server that sent the original
packet, thus creating a loop.
To resolve this problem, all of the switches in the same Ethernet
segment, or EVPN, elect a designated forwarder for that Ethernet
segment/EVPN. This is where Type 4 comes in as it helps with the DF
election process and adds a new split horizon rule.
The election process consists of a series of switches in the same Ethernet segment (ES) creating a list of the all of the IP addresses of all the
switches in the same ES. Each switch is given a value starting from 0,
which is for the switch with the lowest IP address. Once a DF is
selected, this switch or DF can forward BUM traffic. Other switches in
the same ES that are not elected as the DF will drop BUM traffic.

EVPN Type 5 routes

Type 5 routes for EVPN are generally seen in data center interconnect
(DCI) scenarios. For example, lets say you have BMS Server 1 in one
data center that needs to send traffic to BMS Server 2 in another data
center. Now, our VXLAN tunnel needs to traverse the MPLS WAN
network to keep the Layer2 domain between the two server applications.
So Leaf 1 receives the traffic from Server 1. The leaf encapsulates the
Ethernet frames in to VXLAN and sends those packets to the edge

113

114

Day One: Data Center Fundamentals

router or gateway from this data center to the WAN. The edge gateway
(lets say DC1 MX) will have an IRB interface that will act as the
gateway address for Leaf 1.
The MX will strip off the VXLAN header and do a lookup on the
remaining Ethernet packet, which will have a destination for Server 2
in DC2. The edge gateway strips the Ethernet header and routes the
remaining IP packet based on the IP route table related to its IRB
interface. The MX then uses a Type 5 route that it receives from its
opposite number in DC2, and then forwards it over the VXLAN
tunnel between DC1 MX and DC2 MX.
DC2 MX receives the VXLAN encapsulated packet, strips the VXLAN
encapsulation off, routes the IP packet to its IRB interface and re-encapsulates the packet with a Ethernet header with a destination MAC
address of Server 2. It does the MAC address lookup and forwards the
packet over the VXLAN tunnel to the corresponding leaf and then to
Server 2.
So in this scenario, Type 5 allows for inter-data center traffic to be
forwarded over VXLAN tunnels, but allows the Ethernet MAC to be
translated into IP.
Hopefully, these five types of EVPN routing explain how the EVPN
signals in to BGP. The next element to understand is how the gateway
is distributed across several switches.
MORE? Check out the Juniper TechLibrary for more about EVPN and Juniper

switches such as these overviews: https://www.juniper.net/techpubs/

en_US/junos/topics/concept/evpns-overview-ex9200.html; and, https://
www.juniper.net/techpubs/en_US/junos15.1/topics/concept/evpnsoverview.html.

Distributed Layer 3 Gateway

EVPN based fabrics have the ability to distribute Layer 3 gateways
over several switches in the same way a normal Layer 3 gateway would
be done, but for the overlay.
Using the example shown in Figure 9.9, you can see the IP Fabric with
the leaf switches providing Layer 2 VXLAN gateway functions and the
spine layer providing Layer 3 VXLAN gateway functions.

Figure 9.9

Chapter 9: EVPN Protocol

EVPN Distributing Gateways

Server 1 is configured with a default gateway address of 10.0.1.254.

Both Spine A and B have been configured with the same virtual IP
address of 10.0.1.254 and they share the same virtual MAC address of,
say, 00:01:8d:00:01:02. Both Spine A and B will advertise both Type 1
Ethernet Segment IDs (ESI) as well as the Type 2 MAC addresses. The
leaf layer switch will see equal-cost reachability to the same MAC and
ESI, and as such, will load balance traffic over both links to both
spines.

DC1 and DC2 EVPN Solutions

Okay, how does all this EVPN stuff apply back to the two design
scenarios for DC1 and DC2?
Well, because EVPN is a protocol-based solution, its support is defined
on the switches in our fabric, as is the VXLAN encapsulation. Meaning that for both solutions, the server layer would connect to the leaf
devices (top-of-rack switches or end-of-row chassis) with VLAN
connectivity. Those VLANs would be mapped to the relevant VXLAN
VTEP on the switch ports and then that VTEP is mapped to the
relevant EVPN instance providing that VTEPs Layer 2 domain. As
mentioned before, you can map multiple VTEPs to either a single
EVPN instance, creating a one-to-one mapping, or you can map
multiple VTEPs to a single EVPN instance, creating a many-to-one
mapping.

115

116

Day One: Data Center Fundamentals

This provides you with the flexibility to map related applications and
their associated VLANs together under a single EVPN instance to
provide them with isolated Layer 2 connectivity. You can then replicate
it for other application groups and in time allow for the introduction of
tenanted-based solutions.
One aspect not touched upon is the management of this EVPN-VXLAN
solution. Because its protocol-based the configuration of these elements
need to be managed. This can be done very easily through the use of
Junipers Junos Space Network Director platform, which provides IP
fabric deployment and EVPN-VXLAN configuration and management.
NOTE

Junos Space Network Director is beyond the scope of this book, but
look for its inclusion in a future Day One: Data Center Management
Fundamentals.
The other benefit of EVPN not covered is the extension of Layer 2
between data centers, referred to as data center interconnect (DCI).
While not covered in this book, it has been extensively covered in Day
One: Using Ethernet VPNs for Data Center Interconnect, so please
refer to this excellent Day One book at http://www.juniper.net/us/en/
training/jnbooks/day-one/proof-concept-labs/using-ethernet-vpns/.

MORE?

For a really excellent overview of EVPN, refer to Chapter 6 in The

QFX10000 Series, by Doug Hanks, and published in 2016 by OReilly
Media. See: http://www.juniper.net/us/en/training/jnbooks/oreilly-juniper-library/qfx10000-series/.
What to read next? Try visiting Junipers Network Design and Architecture Center for Data Center Networks for an incredible amount of
links, books, guides, and movies: http://www.juniper.net/documentation/en_US/design-and-architecture/index.html.

Summary
This has been a book about data center fundamentals and how Juniper
Networks technology can build data centers. It glosses over very
complicated topics and issues to give you a fundamental understanding
so you can get started on day one.
Everything in this book is subject to change. Please make every effort
to visit the Juniper links provided throughout these pages for the most
up-to-date data sheets and specifications, as those details will change
much faster than this books ability to track them.
Juniper Networks is the home of many data center and data center
interconnect solutions. This book attempts to favor none while trying
to explain them all. Keep in mind that data center networking is still a
complicated engineering science and no one quite agrees on the perfect
data center even the technical reviewers who helped proof this book
had conflicting advice.
While the book has shown you the basics of how to build a data center,
it stops very abruptly at an architected example, and it says nothing
about data center administration, where CoS, analytics, automation,
and orchestration rule. The author hopes to write that next book for
this Day One series as soon as possible. Until that time, please use the
Juniper website to track the many data center administration tools,
from Juniper Contrail, to the NorthStar Controller, to Juniper cloudbased security products and analytics.

Shadow Work Journal Free PDF
91% (164)
Shadow Work Journal Free PDF
30 pages
Hourglass Workout Program by Luisagiuliet 2
76% (21)
Hourglass Workout Program by Luisagiuliet 2
51 pages
Book of Wisdom 2
87% (167)
Book of Wisdom 2
70 pages
BLING A Novel by Erica Kennedy
100% (19)
BLING A Novel by Erica Kennedy
468 pages
Book of Wisdom Revival of Wisdom
89% (93)
Book of Wisdom Revival of Wisdom
72 pages
12 Week Program: Summer Body Starts Now
87% (46)
12 Week Program: Summer Body Starts Now
70 pages
Book of Wisdom - Unlocked
96% (161)
Book of Wisdom - Unlocked
123 pages
Knee Ability Zero Now Complete As A Picture Book 4 PDF Free
94% (68)
Knee Ability Zero Now Complete As A Picture Book 4 PDF Free
49 pages
Aussie Fitness - Low Calorie High Protein Recipes (BOOTLEG)
100% (12)
Aussie Fitness - Low Calorie High Protein Recipes (BOOTLEG)
462 pages
The Hold Me Tight Workbook - Dr. Sue Johnson
100% (16)
The Hold Me Tight Workbook - Dr. Sue Johnson
187 pages
GB Level Up Guide 3 Elevate PDF
79% (28)
GB Level Up Guide 3 Elevate PDF
115 pages
Read People Like A Book by Patrick King-Edited
58% (81)
Read People Like A Book by Patrick King-Edited
12 pages
Livingood, Blake - Livingood Daily Your 21-Day Guide To Experience Real Health
77% (13)
Livingood, Blake - Livingood Daily Your 21-Day Guide To Experience Real Health
260 pages
The Subtle Art of Not Giving a F*ck: A Counterintuitive Approach to Living a Good Life
From Everand
The Subtle Art of Not Giving a F*ck: A Counterintuitive Approach to Living a Good Life
Mark Manson
4/5 (6390)
Cheat Code To The Universe
94% (79)
Cheat Code To The Universe
34 pages
Facial Gains Guide (001 081)
91% (45)
Facial Gains Guide (001 081)
81 pages
Curse of Strahd
95% (467)
Curse of Strahd
258 pages
The Psychiatric Interview - Daniel Carlat
91% (34)
The Psychiatric Interview - Daniel Carlat
473 pages
The Crystal Bible - Judy Hall PDF
94% (185)
The Crystal Bible - Judy Hall PDF
401 pages
The Ultimate Prepared Bug in Guide
100% (5)
The Ultimate Prepared Bug in Guide
149 pages
Can't Hurt Me
69% (29)
Can't Hurt Me
10 pages
The Borax Conspiracy
91% (57)
The Borax Conspiracy
14 pages
Carl Hart, Charles Ksir - Drugs, Society, and Human Behavior (2015)
100% (17)
Carl Hart, Charles Ksir - Drugs, Society, and Human Behavior (2015)
495 pages
The Manual of Free Energy Devices and Systems (1991)
93% (208)
The Manual of Free Energy Devices and Systems (1991)
128 pages
Couples Therapy Workbook
95% (44)
Couples Therapy Workbook
157 pages
Combs Indictment
91% (11)
Combs Indictment
14 pages
The Hidden Truth of Wealth - Oliver Mercer
100% (10)
The Hidden Truth of Wealth - Oliver Mercer
46 pages
D&D Player Handbook 5e
96% (337)
D&D Player Handbook 5e
293 pages
Shortcut To Shred Ebook Revised 9-9-2015 PDF
88% (8)
Shortcut To Shred Ebook Revised 9-9-2015 PDF
15 pages
Workbook For Adult Children of Emotionally Immature Parents - How To Heal From Distant, Rejecting, or Self-Involved Parents
88% (16)
Workbook For Adult Children of Emotionally Immature Parents - How To Heal From Distant, Rejecting, or Self-Involved Parents
77 pages
Trauma-Focused ACT - Russ Harris
95% (39)
Trauma-Focused ACT - Russ Harris
568 pages
34 Mystery of The Seven Keys Guide 1.0
100% (3)
34 Mystery of The Seven Keys Guide 1.0
108 pages
Anastasia: The New Broadway Musical (LIBRETTO)
94% (175)
Anastasia: The New Broadway Musical (LIBRETTO)
117 pages
COSMIC CONSCIOUSNESS OF HUMANITY - PROBLEMS OF NEW COSMOGONY (V.P.Kaznacheev,. Л. V. Trofimov.)
94% (214)
COSMIC CONSCIOUSNESS OF HUMANITY - PROBLEMS OF NEW COSMOGONY (V.P.Kaznacheev,. Л. V. Trofimov.)
212 pages
Raisin in The Sun
97% (34)
Raisin in The Sun
92 pages
I Hate You - Don't Leave Me
80% (54)
I Hate You - Don't Leave Me
6 pages
TDA Birth Certificate Bond Instructions
97% (284)
TDA Birth Certificate Bond Instructions
4 pages
Principles: Life and Work
From Everand
Principles: Life and Work
Ray Dalio
4/5 (639)
The Secret Language of Attraction
86% (107)
The Secret Language of Attraction
278 pages
How To Develop and Write A Grant Proposal
83% (542)
How To Develop and Write A Grant Proposal
17 pages
The Gifts of Imperfection: Let Go of Who You Think You're Supposed to Be and Embrace Who You Are
From Everand
The Gifts of Imperfection: Let Go of Who You Think You're Supposed to Be and Embrace Who You Are
Brené Brown
4/5 (1173)
Supreme Mathematics
80% (25)
Supreme Mathematics
1 page
The Definitive Guide To Project 2025Fnl Project 2025
90% (10)
The Definitive Guide To Project 2025Fnl Project 2025
67 pages
Penis Enlargement Secret
60% (124)
Penis Enlargement Secret
12 pages
Workbook For The Body Keeps The Score
89% (53)
Workbook For The Body Keeps The Score
111 pages
CPN Tut
95% (20)
CPN Tut
23 pages
Never Split the Difference: Negotiating As If Your Life Depended On It
From Everand
Never Split the Difference: Negotiating As If Your Life Depended On It
Chris Voss
4.5/5 (983)
Donald Trump & Jeffrey Epstein Rape Lawsuit and Affidavits
83% (1016)
Donald Trump & Jeffrey Epstein Rape Lawsuit and Affidavits
13 pages
10DDStarterKit PDF
50% (2)
10DDStarterKit PDF
9 pages
KamaSutra Positions
78% (69)
KamaSutra Positions
55 pages
KamaSutra Positions
68% (84)
KamaSutra Positions
55 pages
Organizing Solutions For People With ADHD, 2nd Edition-Revised and Updated
95% (58)
Organizing Solutions For People With ADHD, 2nd Edition-Revised and Updated
221 pages
7 Hermetic Principles
93% (30)
7 Hermetic Principles
3 pages
27 Feedback Mechanisms Pogil Key
77% (13)
27 Feedback Mechanisms Pogil Key
6 pages
The Glass Castle: A Memoir
From Everand
The Glass Castle: A Memoir
Jeannette Walls
4.5/5 (1847)
Frank Hammond - List of Demons
92% (92)
Frank Hammond - List of Demons
3 pages
Grit: The Power of Passion and Perseverance
From Everand
Grit: The Power of Passion and Perseverance
Angela Duckworth
4/5 (650)
Phone Codes
79% (28)
Phone Codes
5 pages
Bessel Van Der Kolk - The Body Keeps The Score - Brain, Mind, and Body in The Healing of Trauma-Penguin (2014)
100% (12)
Bessel Van Der Kolk - The Body Keeps The Score - Brain, Mind, and Body in The Healing of Trauma-Penguin (2014)
490 pages
36 Questions That Lead To Love
91% (35)
36 Questions That Lead To Love
3 pages
Sample Mental Health Progress Note
96% (47)
Sample Mental Health Progress Note
3 pages
How 2 Setup Trust
97% (307)
How 2 Setup Trust
3 pages
2025 MandateForLeadership FULL
70% (10)
2025 MandateForLeadership FULL
920 pages
Jeffrey Epstein's Little Black Book
83% (83)
Jeffrey Epstein's Little Black Book
91 pages
Shoe Dog: A Memoir by the Creator of Nike
From Everand
Shoe Dog: A Memoir by the Creator of Nike
Phil Knight
4.5/5 (627)
How To Kiss A Woman's Breast
60% (114)
How To Kiss A Woman's Breast
14 pages
House of Leaves - Mark Z Danielewski
81% (21)
House of Leaves - Mark Z Danielewski
750 pages
The Hard Thing About Hard Things: Building a Business When There Are No Easy Answers
From Everand
The Hard Thing About Hard Things: Building a Business When There Are No Easy Answers
Ben Horowitz
4.5/5 (361)
Hidden Figures: The American Dream and the Untold Story of the Black Women Mathematicians Who Helped Win the Space Race
From Everand
Hidden Figures: The American Dream and the Untold Story of the Black Women Mathematicians Who Helped Win the Space Race
Margot Lee Shetterly
4/5 (1007)
(Psilocybin) How To Grow Magic Mushrooms A Simple Psilocybe Cubensis Growing Technique PDF
75% (8)
(Psilocybin) How To Grow Magic Mushrooms A Simple Psilocybe Cubensis Growing Technique PDF
48 pages
Singer's Anthology Master Song and Show Index 2008 PDF
37% (43)
Singer's Anthology Master Song and Show Index 2008 PDF
38 pages
Cellular Communication POGIL
80% (10)
Cellular Communication POGIL
5 pages
Lundy - Why Does He Do That PDF
79% (42)
Lundy - Why Does He Do That PDF
1,020 pages
36 Questions To Fall in Love 1
97% (32)
36 Questions To Fall in Love 1
2 pages
Steve Jobs
From Everand
Steve Jobs
Walter Isaacson
4.5/5 (1138)
The 36 Questions That Lead To Love - The New York Times
94% (34)
The 36 Questions That Lead To Love - The New York Times
3 pages
100 Questions To Ask Your Partner
80% (35)
100 Questions To Ask Your Partner
2 pages
Encyclopedia of Herbal Medicine - 3rd Edition (DK Publishing) (2016)
99% (163)
Encyclopedia of Herbal Medicine - 3rd Edition (DK Publishing) (2016)
338 pages
25 Most Valuable US Pennies
94% (16)
25 Most Valuable US Pennies
4 pages
48 Laws of Power Cheat Sheet
97% (30)
48 Laws of Power Cheat Sheet
6 pages
Elon Musk: Tesla, SpaceX, and the Quest for a Fantastic Future
From Everand
Elon Musk: Tesla, SpaceX, and the Quest for a Fantastic Future
Ashlee Vance
4.5/5 (581)
The Emperor of All Maladies: A Biography of Cancer
From Everand
The Emperor of All Maladies: A Biography of Cancer
Siddhartha Mukherjee
4.5/5 (297)
Satanic Calendar
25% (56)
Satanic Calendar
4 pages
DBT Assignment Workbook F0220
100% (65)
DBT Assignment Workbook F0220
218 pages
How To Write A GREAT Business Plan
87% (245)
How To Write A GREAT Business Plan
31 pages
The Startup Guide - Create A Business Plan
88% (199)
The Startup Guide - Create A Business Plan
26 pages
Do You Like Big Girls V01
19% (27)
Do You Like Big Girls V01
161 pages
The 36 Questions That Lead To Love - The New York Times
95% (21)
The 36 Questions That Lead To Love - The New York Times
3 pages
7 Philosophies of Education
75% (8)
7 Philosophies of Education
8 pages
A Man Called Ove: A Novel
From Everand
A Man Called Ove: A Novel
Fredrik Backman
4.5/5 (5141)
14 Easiest & Hardest Muscles To Build (Ranked With Solutions)
100% (8)
14 Easiest & Hardest Muscles To Build (Ranked With Solutions)
27 pages
Jeffrey Epstein39s Little Black Book Unredacted PDF
75% (12)
Jeffrey Epstein39s Little Black Book Unredacted PDF
95 pages
Chart of Seasons and Months Around The World
83% (35)
Chart of Seasons and Months Around The World
1 page
Angela's Ashes: A Memoir
From Everand
Angela's Ashes: A Memoir
Frank McCourt
4.5/5 (943)
The Yellow House: A Memoir (2019 National Book Award Winner)
From Everand
The Yellow House: A Memoir (2019 National Book Award Winner)
Sarah M. Broom
4/5 (100)
The Little Book of Hygge: Danish Secrets to Happy Living
From Everand
The Little Book of Hygge: Danish Secrets to Happy Living
Meik Wiking
3.5/5 (458)
Music Theory - From Beginner To Expert - The Ultimate Step-By-Step Guide To Understanding and Learning Music Theory Effortlessly (Essential Learning Tools For Musicians Book 1)
95% (44)
Music Theory - From Beginner To Expert - The Ultimate Step-By-Step Guide To Understanding and Learning Music Theory Effortlessly (Essential Learning Tools For Musicians Book 1)
145 pages
Yes Please
From Everand
Yes Please
Amy Poehler
4/5 (1994)
Devil in the Grove: Thurgood Marshall, the Groveland Boys, and the Dawn of a New America
From Everand
Devil in the Grove: Thurgood Marshall, the Groveland Boys, and the Dawn of a New America
Gilbert King
4.5/5 (278)
The World Is Flat 3.0: A Brief History of the Twenty-first Century
From Everand
The World Is Flat 3.0: A Brief History of the Twenty-first Century
Thomas L. Friedman
3.5/5 (2283)
How To Come Up With Good Ideas For Startups - The Scribd Story and The Trip Method
87% (487)
How To Come Up With Good Ideas For Startups - The Scribd Story and The Trip Method
22 pages
Bad Feminist: Essays
From Everand
Bad Feminist: Essays
Roxane Gay
4/5 (1077)
Independent Contractor Agreements
80% (839)
Independent Contractor Agreements
3 pages
The Outsider: A Novel
From Everand
The Outsider: A Novel
Stephen King
4/5 (2876)
ALCHEMIST
64% (14)
ALCHEMIST
4 pages
1001 Songs
69% (72)
1001 Songs
1,798 pages
Trademark License Agreement
79% (382)
Trademark License Agreement
3 pages
The Sympathizer: A Novel (Pulitzer Prize for Fiction)
From Everand
The Sympathizer: A Novel (Pulitzer Prize for Fiction)
Viet Thanh Nguyen
4.5/5 (141)
The 4 Hour Workweek, Expanded and Updated by Timothy Ferriss - Excerpt
23% (954)
The 4 Hour Workweek, Expanded and Updated by Timothy Ferriss - Excerpt
38 pages
Zodiac Sign & Their Most Common Addictions
63% (30)
Zodiac Sign & Their Most Common Addictions
9 pages
A Heartbreaking Work Of Staggering Genius: A Memoir Based on a True Story
From Everand
A Heartbreaking Work Of Staggering Genius: A Memoir Based on a True Story
Dave Eggers
3.5/5 (233)
Team of Rivals: The Political Genius of Abraham Lincoln
From Everand
Team of Rivals: The Political Genius of Abraham Lincoln
Doris Kearns Goodwin
4.5/5 (244)
On Fire: The (Burning) Case for a Green New Deal
From Everand
On Fire: The (Burning) Case for a Green New Deal
Naomi Klein
4/5 (78)
Fear: Trump in the White House
From Everand
Fear: Trump in the White House
Bob Woodward
3.5/5 (835)
Rise of ISIS: A Threat We Can't Ignore
From Everand
Rise of ISIS: A Threat We Can't Ignore
Jay Sekulow
3.5/5 (144)
John Adams
From Everand
John Adams
David McCullough
4.5/5 (2546)
The Light Between Oceans: A Novel
From Everand
The Light Between Oceans: A Novel
M.L. Stedman
4.5/5 (813)
Noncompete Agreement
87% (141)
Noncompete Agreement
2 pages
Best of Best Women's Erotica by Marcy Sheiner PDF
41% (17)
Best of Best Women's Erotica by Marcy Sheiner PDF
209 pages
The Unwinding: An Inner History of the New America
From Everand
The Unwinding: An Inner History of the New America
George Packer
4/5 (45)

DC Fundamentals

Uploaded by

DC Fundamentals

Uploaded by

Fundamentals Series

Day One: Data Center

Learn all the basics of data centers,

DAY ONE: Data Center FUNDAMENTALS

Day One: Data Center Fundamentals

Chapter 1: Common Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2016 by Juniper Networks, Inc. All rights reserved.

About the Author

Welcome to Day One

Search for Juniper Networks Books.

(Android, Kindle, iPad, PC, or Mac) by opening your devices

vervante.com) or Amazon (amazon.com) for between $12-$28,

Audience and Scope of This Book

What You Need to Know Before Reading This Book

What You Will Learn by Reading This Book

scale and operations

looks at the different architectures you can use, depending on

larger design, but it can have interesting repercussions if not

about when designing a network. If you know the speeds at which

 Chapter Five: This fabric architecture chapter looks at different

solutions for the connectivity of products, their management,

switches talk and transport traffic to each other. This connects to

types of overlays supported, how they interact with the underlay

networks examines the benefits of using either a single pane of

solution for overlay networks through the use of Ethernet VPN

Switches Used In the Data Center

Download each products datasheet for up-to-date improvements

Day One: Data Center Fundamentals

The QFX5100 Series

The QFX5100-48T is a tri-speed 100/1000/10Gb BASE-T top-of-rack

Chapter 1: Common Components

The QFX5100-24Q is a 40-Gigabit Ethernet QSFP+ switch with 24

Day One: Data Center Fundamentals

The QFX5100-24Q-AA is the same as the QFX5100-24Q with 24

Chapter 1: Common Components

The QFX5200 Series

The QFX5200-32C is a 100Gigabit Ethernet top-of-rack switch that

Day One: Data Center Fundamentals

The QFX10000 Series

The QFX10002-36Q is a 100Gigabit Ethernet aggregation and spine

The QFX10002-36Q comes standard with redundant fans and

The QFX10002-72Q is a 100Gigabit Ethernet aggregation and spine

The QFX10002-72Q comes standard with redundant fans and

Chapter 1: Common Components

Complete, up-to-date details on the two fixed QFX10000 Series

The QFX10008 is a chassis-based high-density aggregation and spine

The QFX10008 supports three line card options:

ports of 100GbE, and 144 ports of 10GbE with breakout cables

with six ports of 40GbE and 2 ports of 100GbE

The QFX10016 is a chassis-based aggregation and spine layer switch

Day One: Data Center Fundamentals

The QFX10016 supports three line card options:

ports of 100GbE, and 144 ports of 10GbE

six ports of 40GbE, and two ports of 100GbE

Up-to-date details on the chassis-based QFX10000 series switches can

Chapter 1: Common Components

Custom and Merchant Silicon

Day One: Data Center Fundamentals

The most important aspect in this requirement is making sure the

Now that you have a better understanding of the Juniper products

Day One: Data Center Fundamentals

ToR Switch Positions

The switches deployed at the top-of-rack are typically of a fixed

Connectivity from the top-of-rack switch to the aggregation layer is

ToR to Aggregation Uplinks

For uplink cables, fiber provides a considerably better investment for

Day One: Data Center Fundamentals

the configuration can be duplicated for the majority of switches, that

Juniper has addressed these concerns by providing virtualization

ToR Cross Connect

Summary of ToR Design

Cabling complexity is reduced as all the servers are

There may be more unused ports in each rack and it

If the racks are small, there could be one network

Unplanned Expansions (within a rack) might be

ToR architecture supports modular deployment of

Chapter Five: This fabric architecture chapter looks at different