Scribd
Upload
61 views

Introduction To Web Application Load Testing

This document provides an introduction to web application load testing. It discusses concepts like profiling, load testing, and monitoring. Load testing aims to measure an application's ability to handle simultaneous users and transactions with adequate response times. The document outlines different types of load testing and tools for performing load testing, like JMeter. It also discusses common web application performance issues and the typical order for addressing them, such as improving application design, upgrading hardware/software, or changing architectures.

Uploaded by

nagpathuri
Copyright
© Attribution Non-Commercial (BY-NC)
Available Formats
Download as PDF, TXT or read online on Scribd
61 views

Introduction To Web Application Load Testing

This document provides an introduction to web application load testing. It discusses concepts like profiling, load testing, and monitoring. Load testing aims to measure an application's ability to handle simultaneous users and transactions with adequate response times. The document outlines different types of load testing and tools for performing load testing, like JMeter. It also discusses common web application performance issues and the typical order for addressing them, such as improving application design, upgrading hardware/software, or changing architectures.

Uploaded by

nagpathuri
Copyright
© Attribution Non-Commercial (BY-NC)
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 14

Introduction to Web Application Load Testing

Philip Johnson Collaborative Software Development Laboratory University of Hawaii

(1)

Concepts and terms

(2)

Testing, Profiling, Load Testing


Functional testing: Determine whether application implements desired capabilities correctly. Non-functional testing: Determine whether functionally correct program provides adequate responsiveness.

Profiling: Examine a single software application CPU and s memory usage. Result: Detect most significant performance problem in software application and remove it.
(3)

Load testing: Measure the applications ability to sustain a number of simultaneous users and/or transactions with adequate response times. Result: changes to hardware, software infrastructure, or application design

Active vs. Passive Monitoring


Active Monitoring: Generate traffic data from virtual users. Provides early warning of problems Measures against a baseline Enables pro-active problem solution before users impacted Passive monitoring: Collect traffic data from real users Provides actual level of users and their locations Non-intrusive Helps prioritize issues

You want both active and passive monitoring.


(4)

Flavors of Load Testing


Load testing: Test system against a specified number of users. Issue: can system handle users correctly with acceptable responsiveness? Stress testing: Test system against specified number of users over long period of time. Issue: is system stable and reliable under sustained load? Capacity testing: Test system against an increasing number of users. Issue: at what load does system begin failing to handle requests correctly and with acceptable

(5)

Warning!
Capacity load testing is a denial-of-service attack! When testing across network, you must: Think through impact on network and other users of performing test Notify and obtain permission from network administrators or ISPs Schedule test for off-hours Failure to observe appropriate netiquette can lead to reprisals. Do NOT: Capacity load test using katrina or ICS lab machines!

(6)

My web application is slow. Should I: Add RAM? Move from Windows to Linux? Optimize my application code? Allocate more memory to the JVM? Move from Tomcat to Resin? To Apache? Move the database to its own server? Change the database indexing? Move from InstantDB to Oracle? Upgrade my Internet bandwidth? Use Enterprise Java Beans?
(7)

Diagnosing web app performance problems is hard

Load Testing Process


1. Determine load testing research question(s): Should we upgrade to new server or software? Where are bottlenecks in current system? How many concurrent users without failure? 2. Determine the system data to collect Response time, CPU usage, RAM usage, Page errors 3. Determine the testing tool to use Basic: JMeter, WebApp Stress Tool Enterprise-class: Mercury Interactive, etc. 4. Determine baseline data values What is the hardware/software configuration of web app? What are load test results under normal conditions? 5. Collect experimental data values Collect load test data with new load, hardware, or software 6. Determine changes (if any) required to system.
(8)

Web Application Architectures

(9)

Simple client-server architecture


Client Web Server, Application System, Database

Internet

(10)

Six variables impact on performance: Internet bandwidth (T-1, 56K modem, etc.) Server hardware configuration (RAM, Disk speed, OS) Web server software (Tomcat vs Resin, etc.) Database server software (InstantDB, db40, etc.) Application design (caching, algorithms, etc.) Database design (indexes, tables, queries)

Simple N-tier architecture


Client Web Server Application Server Database Server

Internet

LAN

LAN

10 variables impact on performance: Internet bandwidth Web server hardware, software, app design App server hardware, software, app design DB server hardware, software, DB design
(11)

Industry-strength N-tier architecture


Client Internet Firewall Switch Web Server Farm Load Balancer

DB Servers

(12)

Don try this at home, kids. t

App Server Farm

Web app symptoms and typical underlying problems

(13)

Symptoms of web app problems


Long user response time Long server response time Memory leaks High CPU usage Too many open connections Length queues for requests Too many table scans of database Database deadlocks Erroneous data returned HTTP errors Pages not available
(14)

Typical database problems


Insufficient indexing Tune database indexing to improve query processing Fragmented databases Place table records into adjacent pages. Out-of-date statistics Degrade query optimizer performance Faulty application design Excessive DB calls, excessive data requests

(15)

Typical Web Server problems


Poor server design Inefficient data or page caching Use of Active Server Pages Memory problems Physical memory constraints High CPU usage Usage > 70% indicates problems

(16)

Typical app server problems


Poor database tuning Application server sending too many DB requests Poor cache management Produces high CPU usage, disk access Poor session management Produces high CPU usage, disk access Time-outs Poor security design Excessive use of https protocol.
(17)

Network problems
Requires specialized testing to uncover (must stress network without causing application/db stress). Potential sources of network problems: Firewall throughput Internet access throughput Load balancers, gateways, routers

(18)

Load testing tools

(19)

Example Load Testing Tools


Enterprise Solutions: Astra LoadTest, Mercury Interactive - $10K for 50 v-users, $30K for 250 v-users RadView WebLoad - $10K for 100 v-users, $38K for 1K v-users Personal Solutions: Microsoft Web Application Stress Tool - Free from webtool.rte.microsoft.com - Windows-specific JMeter - Open source - Cross-platform We will look at JMeter
(20)

A load testing scenario


1. Determine load testing research question(s): What is the baseline performance of EmailFinder? 2. Determine the system data to collect User response time Server CPU and memory usage Web app memory usage Any failures 3. Determine the testing tool to use JMeter 4. Determine baseline data values Server hardware: PIII/700, 384m RAM, FastEthernet LAN Let look at the JMeter configuration required to s provide more baseline data.

(21)

Downloading, Installation, Invocation


Go to http://jakarta.apache.org/jmeter/ Download and expand the JMeter application zip file. Change directory to jakarta-jmeter/bin Invoke jmeter.bat (or jmeter on unix). Don forget to RTFM: t Online documentation www.mail-archive.com/[email protected] Documentation is not spectacular.

(22)

Initial screen

(23)

Top-level Thread Group

(24)

Default Page Values

(25)

Invoke login servlet

(26)

Specify session cookie

(27)

Invoke finger servlet

(28)

Specify 1 sec between requests

(29)

Response time with 1 thread

(30)

10

Server stats, 1 thread

(31)

Response time with 5 threads

(32)

Server stats, 5 threads

(Representative)

(33)

11

Response time with 10 threads

(34)

Server stats, 10 threads

(Representative)

(35)

Conclusions
With 10 threads at rate of 1 page/second, Response time averages 5 seconds. Server CPU utilization at around 30% Web server process requires 25MB. Load test methodology problems: Web server used for other web apps Did not vary search term, session ID. Stress testing (12-24 hr) not performed Capacity testing not performed. No checking for correct page response as load increased. Next steps: Check server under stress testing conditions. Fix methodology problems. Repeat with alternative hardware, software.
(36)

12

JMeter Tips
Use 1.6.1 for now. (I had problems w/1.7b) Make sure you load your .jmx file inside of Test Plan, otherwise it isn executed. t Possible additions to jmeter.bat java cmd: s -Xms64m Xmx128m (increase memory) -Dsun.java2d.noddraw (eliminate GUI flicker) When entering a field value in GUI, make sure to exit field so that value is seen!
(37)

JMeter pros and cons


Pros: Open source Active development Free Cross-platform Extensible architecture support plug-ins Good project to help work on. Best free alternative I could find. Cons: Documentation marginal How to automate 100 s of separate v-users with separate session IDs? Unclear how to detect failure (I.e. 404s)--maybe assertion?

(38)

Summary

(39)

13

Typical order of fixes


1. Improve current application design. Algorithms, caching, DB calls, memory use 2a. Upgrade hardware RAM, CPU, network bandwidth 2b. Upgrade software infrastructure OS, web server, database 3. Upgrade system architecture Client-server to basic n-tier Basic n-tier to enterprise n-tier Requires software and hardware changes.
(40)

Keep it in perspective
Functional correctness is more important than non-functional correctness. Not every WebApp requires scalability. High volume web sites require professional load testing.

(41)

14

You might also like