Front cover

IBM Technology for

Java Virtual Machine in
IBM i5/OS
Most information about 32-bit JVM
applies to all IBM server platforms

Comprehensive book on new

32-bit JVM in i5/OS

Run WebSphere Application

Server in 32-bit JVM

Aleksandr Nartovich
Adam Smye-Rumsby
Paul Stimets
George Weaver

ibm.com/redbooks
International Technical Support Organization

IBM Technology for Java Virtual Machine in IBM i5/OS

February 2007

SG24-7353-00
 Note: Before using this information and the product it supports, read the information in “Notices” on
page vii.

First Edition (February 2007)

This edition applies to IBM® Developer Kit and Runtime Environment, Java™ 2 Technology Edition, Version
5.0.

© Copyright International Business Machines Corporation 2007. All rights reserved.

Note to U.S. Government Users Restricted Rights -- Use, duplication or disclosure restricted by GSA ADP Schedule
Contract with IBM Corp.
Contents

Notices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
Trademarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
The team that wrote this IBM Redbook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
Become a published author . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .x
Comments welcome. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .x

Chapter 1. The world of Java . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1 Java: language and platform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.1 Java language. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.2 Java platform and Java software development kit . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Java Virtual Machine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.1 JVM architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Java Runtime Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Java platform development tools and API. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

Chapter 2. Exploring IBM Technology for Java Virtual Machine . . . . . . . . . . . . . . . . . . 9

2.1 A history of Classic JVM and IBM Technology for JVM on System i platform . . . . . . . 10
2.1.1 The System i Classic JVM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.1.2 IBM Technology for Java Virtual Machine. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2 Comparing IBM Technology for JVM and Classic JVM. . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.1 Similarities. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.2 Differences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 Choosing IBM Technology for JVM or Classic JVM . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.4 Using IBM Technology for JVM with existing applications . . . . . . . . . . . . . . . . . . . . . . 16
2.5 Future JVM developments on i5/OS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

Chapter 3. New user guide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

3.1 i5/OS portable application solutions environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.2 Installing IBM Technology for JVM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.2.1 Checking prerequisites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.2.2 Installing IBM Technology for JVM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2.3 Verifying the installation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.2.4 Uninstalling IBM Technology for JVM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.3 Basic configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.3.1 Setting JAVA_HOME . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.3.2 Setting default command line options in SystemDefault.properties file . . . . . . . . 25
3.4 Known issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

Chapter 4. Making the switch to IBM Technology for JVM . . . . . . . . . . . . . . . . . . . . . . 27

4.1 Fitting your application into a 32-bit JVM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.1.1 Measuring heap usage in Classic JVM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.2 Configuring garbage collection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.2.1 Choosing the right policy. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.2.2 The minimum and maximum heap size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.3 Finding dependencies to the Classic JVM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.4 WebSphere Application Server and IBM Technology for JVM . . . . . . . . . . . . . . . . . . . 33

© Copyright IBM Corp. 2007. All rights reserved. iii

 Chapter 5. Tuning the garbage collector. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.1 Introduction to garbage collection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
5.1.1 Garbage collection in IBM Technology for JVM . . . . . . . . . . . . . . . . . . . . . . . . . . 36
5.2 Available options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
5.2.1 Garbage collection policies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
5.2.2 Tuning heap size. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5.2.3 Verbose GC output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.3 Choosing the right policy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

Chapter 6. Optimizing performance with shared classes . . . . . . . . . . . . . . . . . . . . . . . 49

6.1 Introduction to shared classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
6.1.1 Classes and classloaders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
6.1.2 Overview of shared classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
6.1.3 History of shared classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
6.1.4 Shared classes and user-defined classloaders. . . . . . . . . . . . . . . . . . . . . . . . . . . 53
6.2 Shared classes in IBM Technology for JVM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
6.2.1 The class cache . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
6.2.2 Deploying shared classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
6.2.3 Recommendations for shared classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
6.3 Shared classes and application performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
6.3.1 When to use shared classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
6.3.2 Overheads associated with shared classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
6.3.3 Performance gains with shared classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
6.4 Shared Classes Helper API . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
6.4.1 Helper API overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
6.4.2 Helper API usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
6.4.3 Shared Classes Helper API example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
6.4.4 Shared Classes Helper API Javadoc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
6.5 Security considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
6.5.1 Operating system security. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
6.5.2 Java security . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
6.6 Bytecode modification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

Chapter 7. Java Native Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

7.1 Java Native Interface overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
7.2 When must you use JNI? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
7.3 JNI considerations in IBM Technology for JVM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
7.3.1 Teraspace storage model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
7.3.2 Thread safety of i5/OS system functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
7.3.3 64-bit PASE native methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
7.3.4 Adopted authority . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
7.3.5 Performance of native code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
7.4 Diagnostic messages and files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
7.5 Further information about the JNI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

Chapter 8. Analyzing JVM behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

8.1 JVM analysis overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
8.2 JVM does not start . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
8.2.1 Quick fix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
8.2.2 Environment problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
8.2.3 JVM problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
8.2.4 Application problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
8.3 JVM crash . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
8.3.1 Causes of a crash . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

iv IBM Technology for Java Virtual Machine in IBM i5/OS

 8.3.2 Quick fix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
8.3.3 If application runs properly under Classic JVM . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
8.3.4 If the Classic JVM also crashes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
8.3.5 Identifying the cause of a crash . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
8.4 Unresponsive application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
8.4.1 Overview of a hang . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
8.4.2 Hangs in applications running on WebSphere Application Server . . . . . . . . . . . . 91
8.4.3 Quick fix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
8.4.4 Troubleshooting a hang . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
8.5 Investigating a suspected memory leak . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
8.5.1 Quick fix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
8.5.2 Memory leaks and the heap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
8.5.3 Collecting heap information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
8.5.4 Using the Memory Dump Diagnostic for Java (MDD4J) tool. . . . . . . . . . . . . . . . 114
8.6 Performance issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
8.6.1 Quick fix: general guidelines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
8.6.2 Server guidelines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
8.6.3 Java virtual machine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
8.6.4 Application guidelines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
8.7 Application debugging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
8.7.1 Review any available data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
8.7.2 Debugging a Java application running on System i platform with WebSphere
Development Studio Client for iSeries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

Appendix A. Running WebSphere Application Server with IBM Technology

for JVM. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
System requirements for i5/OS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
Switching between JVMs in WebSphere Application Server 6.1 . . . . . . . . . . . . . . . . . . . . 168
The enableJVM script . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
Enable an existing WebSphere Application Server profile with IBM Technology for JVM . 170
Determining the JVM in use . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
View the job log. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
Configuring the WebSphere Application Server JVM . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
Monitoring the WebSphere Application Server JVM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
Monitoring with Tivoli Performance Viewer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
Verbose garbage collection and logging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
Additional information. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179

Appendix B. IBM Support Assistant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181

Overview of IBM Support Assistant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
Install IBM Support Assistant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
Prerequisites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
Downloading IBM Support Assistant from Web . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
Configuring and updating IBM Support Assistant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
Uninstall IBM Support Assistant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
Use IBM Support Assistant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
Starting IBM Support Assistant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
Starting JVM tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
Submitting a problem report to IBM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
Troubleshooting IBM Support Assistant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189

Appendix C. Diagnostic Tools and Framework for Java Dumps . . . . . . . . . . . . . . . . 191

Overview of DTFJ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
Tools based on DTFJ API . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193

Contents v
 DTFJ API documentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
DTFJ API or JVMTI? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194

Appendix D. jconsole . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195

Monitoring IBM Technology for JVM remotely with a GUI . . . . . . . . . . . . . . . . . . . . . . . . . 196

Appendix E. Additional material . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199

Locating the Web material . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
Using the Web material . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
How to use the Web material . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199

Related publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201

IBM Redbooks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
Online resources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
How to get IBM Redbooks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
Help from IBM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203

vi IBM Technology for Java Virtual Machine in IBM i5/OS

Notices

This information was developed for products and services offered in the U.S.A.

IBM may not offer the products, services, or features discussed in this document in other countries. Consult
your local IBM representative for information on the products and services currently available in your area. Any
reference to an IBM product, program, or service is not intended to state or imply that only that IBM product,
program, or service may be used. Any functionally equivalent product, program, or service that does not
infringe any IBM intellectual property right may be used instead. However, it is the user's responsibility to
evaluate and verify the operation of any non-IBM product, program, or service.

IBM may have patents or pending patent applications covering subject matter described in this document. The
furnishing of this document does not give you any license to these patents. You can send license inquiries, in
writing, to:
IBM Director of Licensing, IBM Corporation, North Castle Drive, Armonk, NY 10504-1785 U.S.A.

The following paragraph does not apply to the United Kingdom or any other country where such
provisions are inconsistent with local law: INTERNATIONAL BUSINESS MACHINES CORPORATION
PROVIDES THIS PUBLICATION "AS IS" WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESS OR
IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF NON-INFRINGEMENT,
MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Some states do not allow disclaimer of
express or implied warranties in certain transactions, therefore, this statement may not apply to you.

This information could include technical inaccuracies or typographical errors. Changes are periodically made
to the information herein; these changes will be incorporated in new editions of the publication. IBM may make
improvements and/or changes in the product(s) and/or the program(s) described in this publication at any time
without notice.

Any references in this information to non-IBM Web sites are provided for convenience only and do not in any
manner serve as an endorsement of those Web sites. The materials at those Web sites are not part of the
materials for this IBM product and use of those Web sites is at your own risk.

IBM may use or distribute any of the information you supply in any way it believes appropriate without incurring
any obligation to you.

Information concerning non-IBM products was obtained from the suppliers of those products, their published
announcements or other publicly available sources. IBM has not tested those products and cannot confirm the
accuracy of performance, compatibility or any other claims related to non-IBM products. Questions on the
capabilities of non-IBM products should be addressed to the suppliers of those products.

This information contains examples of data and reports used in daily business operations. To illustrate them
as completely as possible, the examples include the names of individuals, companies, brands, and products.
All of these names are fictitious and any similarity to the names and addresses used by an actual business
enterprise is entirely coincidental.

COPYRIGHT LICENSE:

This information contains sample application programs in source language, which illustrate programming
techniques on various operating platforms. You may copy, modify, and distribute these sample programs in
any form without payment to IBM, for the purposes of developing, using, marketing or distributing application
programs conforming to the application programming interface for the operating platform for which the sample
programs are written. These examples have not been thoroughly tested under all conditions. IBM, therefore,
cannot guarantee or imply reliability, serviceability, or function of these programs.

© Copyright IBM Corp. 2007. All rights reserved. vii

Trademarks
The following terms are trademarks of the International Business Machines Corporation in the United States,
other countries, or both:
Redbooks (logo) ™ DB2 Universal Database™ Redbooks™
alphaWorks® DB2® System i™
developerWorks® Integrated Language Environment® System i5™
eServer™ IBM® System p™
iSeries™ Language Environment® System p5™
i5/OS® Lotus® System z™
z/OS® OS/390® System z9™
z9™ OS/400® Tivoli®
AIX® PowerPC® WebSphere®
AS/400® POWER™
Domino® Rational®

The following terms are trademarks of other companies:

Enterprise JavaBeans, EJB, Java, Javadoc, JavaBeans, JavaServer, JavaServer Pages, JDBC, JDK, JMX,
JRE, JSP, JVM, J2EE, J2ME, J2SE, Solaris, Sun, Sun Microsystems, and all Java-based trademarks are
trademarks of Sun Microsystems, Inc. in the United States, other countries, or both.

Internet Explorer, Microsoft, Windows, and the Windows logo are trademarks of Microsoft Corporation in the
United States, other countries, or both.

Intel, Intel logo, Intel Inside logo, and Intel Centrino logo are trademarks or registered trademarks of Intel
Corporation or its subsidiaries in the United States, other countries, or both.

Linux is a trademark of Linus Torvalds in the United States, other countries, or both.

Other company, product, or service names may be trademarks or service marks of others.

viii IBM Technology for Java Virtual Machine in IBM i5/OS

Preface

This IBM® Redbook gives a broad understanding of a new 32-bit Java™ Virtual Machine
(JVM™) in IBM i5/OS®. With the arrival of this new JVM, IBM System i™ platform now
comfortably supports Java and WebSphere® applications on a wide array of different server
models: from entry size boxes to the huge enterprise systems.

This book provides in-depth information about setting Java and IBM WebSphere
environments with new 32-bit JVM, tuning its performance, and monitoring or troubleshooting
its runtime with the new set of tools.

Information in this IBM Redbook helps system architects, Java application developers, and
system administrators in their work with 32-bit JVM in i5/OS.

Important: Despite the fact that this book targets i5/OS implementation, most information
in this book applies to all IBM server platforms, where the new 32-bit JVM is supported.

The team that wrote this IBM Redbook

This IBM Redbook was produced by a team of specialists from around the world working at
the International Technical Support Organization, Rochester Center.

Aleksandr V. Nartovich is a Senior IT Specialist in the ITSO, Rochester Center. He joined

the ITSO in January 2001 after working as a developer in the IBM WebSphere Business
Components organization. During the first part of his career, Aleksandr was a developer in
IBM AS/400® communications. Later, he shifted his focus to business components
development on WebSphere. Aleksandr holds two degrees, one in computer science from
the University of Missouri-Kansas City and the other in electrical engineering from Minsk
Radio Engineering Institute.

Adam Smye-Rumsby joined IBM in 2000. He currently works as a Software Support

Specialist for IBM in the UK, solving customer problems related to IBM WebSphere Portal
and Lotus® Domino®. Previously he worked at IBM Hursley Laboratory in the UK testing IBM
Java software development kits (SDKs) on a variety of platforms. He has also worked as a
tester on IBM Web Services technology. Adam holds a Computer Science degree from the
University of Portsmouth.

Paul Stimets is an Advisory Software Engineer at IBM, working in the Rochester Support
Center. He supports Java and WebSphere Application Server on the System i platform. He
has over 8 years of experience at IBM and has worked with Java on System i platform since
its introduction in 1998 with IBM OS/400® V4R2.

George Weaver is a software engineer with the System i Technology Center at IBM in
Rochester, MN. He has provided consultations, education, writings, Web content, and videos
to help System i clients, software vendors, business partners and solution providers enhance
their applications and services portfolios over the last 13 years. His current areas of focus are
Web application development, and WebSphere Application Server configuration, and
administration, and performance.

Many thanks to Java developers at IBM Rochester, MN for their enthusiastic support of this
IBM Redbook.

© Copyright IBM Corp. 2007. All rights reserved. ix

 Thanks to the following people for their contributions to this project:

Bill Berg
Marc Blais
Arv Fisher
Steve Fullerton
Jesse Gorzinski
Sandra Marquardt
Mark Schleusner
Nishant Thakkar
Blair Wyman
IBM Rochester, Minnesota

Richard Chamberlain
Ben Corrie
Holly Cummins
IBM Hursley, UK

Become a published author

Join us for a two-week to six-week residency program! Help write an IBM Redbook dealing
with specific products or solutions, while getting hands-on experience with leading-edge
technologies. You will have the opportunity to team with IBM technical professionals,
Business Partners, and Clients.

Your efforts helps increase product acceptance and customer satisfaction. As a bonus, you
will develop a network of contacts in IBM development labs, and increase your productivity
and marketability.

Find out more about the residency program, browse the residency index, and apply online at:
ibm.com/redbooks/residencies.html

Comments welcome
Your comments are important to us!

We want our IBM Redbooks™ to be as helpful as possible. Send us your comments about
this or other IBM Redbooks in one of the following ways:
򐂰 Use the online Contact us review IBM Redbook form found at:
ibm.com/redbooks
򐂰 Send your comments in an email to:
[email protected]
򐂰 Mail your comments to:
IBM Corporation, International Technical Support Organization
Dept. HYTD Mail Station P099
2455 South Road
Poughkeepsie, NY 12601-5400

x IBM Technology for Java Virtual Machine in IBM i5/OS

 1

Chapter 1. The world of Java

This chapter provides a technical overview of the world of Java. It describes the typical
architecture of the Java platform and its components. Understanding the Java platform and
the functionality of its components is the key to understanding the rest of the materials in this
book.

This chapter includes the following sections:

򐂰 1.1, “Java: language and platform” on page 2
򐂰 1.2, “Java Virtual Machine” on page 4
򐂰 1.3, “Java Runtime Environment” on page 7
򐂰 1.4, “Java platform development tools and API” on page 8

© Copyright IBM Corp. 2007. All rights reserved. 1

1.1 Java: language and platform
Java has become one of the buzz words of the IT world. Not mentioning Java in the list of
skills or in your business presentation is a bad form. Quite often when people say “Java” they
mean many different things. And if you ask them to define more accurately “Java” in the
context of their speech, not all of them give you the correct answer.

Because this is a technical book, it provides a brief introduction to several key definitions
related to Java.

1.1.1 Java language

Java language is one of the most popular programming languages today. It has been around
for many years since Sun™ Microsystems™ has introduced it in early 1990s.

Any programming language is based on a well defined syntax and semantic1. Any high-level
language program, including Java, is a text file with a unique file extension (.java for the Java
programs). You cannot execute a program on a computer as a text file. There is a
requirement for another component, called compiler, that converts human readable text file
into machine instructions. Typically, these instructions are unique for a specific hardware
platform, such as Intel® or IBM PowerPC®. There is a compiler for Java too, but it works
differently from other compilers. Refer to 1.2, “Java Virtual Machine” on page 4 and 1.4, “Java
platform development tools and API” on page 8 for further information.

With time, people have noticed that some programming segments are duplicated in many
different applications. As the result, the most frequently used functions (segments of
programs) are prewritten and grouped in special libraries. These libraries are called
differently for different platforms and languages. For Java they are called Java class libraries.
You can download these packages to a computer and include (or reference) these packages
to any program. Therefore you achieve code reuse.

As a language (meaning syntax and semantic) Java has not changed much, but the runtime
support and the number of available packages have grown tremendously. As a result, Java
programming language is one of the most popular languages for enterprise applications.

The main characteristics of the Java language includes the following:

򐂰 Java is a high-level programming language.
򐂰 Java is an object-oriented language.
Many programming languages support this concept. It is based on the idea that the
program is seen as the collection of objects. Each object represents an entity within the
business task that this program has to solve. For example, many applications have
Customer or Address objects. The objects communicate through well-defined interfaces,
called methods, by exchanging messages.
The object-oriented programming claims to achieve a better flexibility, reusability of the
code, and maintainability. It is a popular choice for the large-scale solutions.
򐂰 Java is platform independent.
A Java program written according to the best practices can be moved to any supported
platform without any change or requirement to recompile.

1 Semantic is the correct set of actions (or machine instructions) executed for each operand in your program.

2 IBM Technology for Java Virtual Machine in IBM i5/OS

1.1.2 Java platform and Java software development kit
To support the development and deployment of the Java applications, you must have a
special environment on your system. This environment is called Java platform. Figure 1-1
shows the architecture of the Java platform. This chapter explores this architecture in more
details in the following sections.

Java SDK

Development tools Tools

(compiler, debugger,
APIs
archiver, etc.)

JRE
Supporting files Core classes

JVM
Interpreter
Memory management
Class loading JIT
Diagnostic

Figure 1-1 Java platform architecture

Different applications might require different Java platforms. Therefore, there are several
editions of the Java platform in existence today:
򐂰 Java Platform Micro Edition (Java ME, formerly called J2ME™): This edition targets
development and deployment of the Java applications for small, mobile devices, such as
PDAs and cell phones.
򐂰 Java Platform Standard Edition (Java SE, formerly J2SE™): This edition targets
development and deployment for most of the Java applications running on the servers.
򐂰 Java Platform Enterprise Edition (Java EE, formerly J2EE™): This edition has additional
support application programming interface (API) for enterprise-level applications. The
examples of additional API include Web services, distributed deployment of the
applications, and communication API.

Java platform is a document which is formed of a set of specifications. These specifications

define the services and APIs for the Java environment. The product, that implements this
document, is called Java software development kit (Java SDK). It includes the following major
components (see Figure 1-1):
򐂰 Java Virtual Machine (JVM)
򐂰 Java Runtime Environment (JRE™)
򐂰 Development tools

Chapter 1. The world of Java 3

1.2 Java Virtual Machine
After you create a Java program, you have to compile this program. The result of this
compilation is the file with extension .class. It contains Java bytecode. The generated
bytecode is identical no matter on what platform you have compiled your program. It is
platform neutral. That is why we can say that Java program is platform independent.

However, you cannot execute this bytecode yet because your system understands machine
instructions that are unique to your system’s architecture (PowerPC, Intel, and so on). In
order to execute a Java program, you have to install a middleware component that fits
between your Java program and operating system. This component is called Java Virtual
Machine. JVM runs as an application on top of an operating system, as shown in Figure 1-2,
and provides a runtime environment for all Java applications. JVM effectively emulates an
operating system environment for Java programs. That is why it is called a Virtual Machine.

Java
application

Java Virtual Machine (JVM)

Operating system
. (i5/OS, AIX, Linux, Windows, etc.)

HW platform
(PowerPC, Intel, etc.)
Figure 1-2 Java Virtual Machine

The main purpose of JVM is to convert the Java bytecode to the machine instructions that
you can execute on the hardware platform where you run your application. JVM is the
overhead in a Java Runtime Environment compared with traditional programming languages,
such as C or RPG. However, this is the price you pay for Java portability.

Converting Java bytecode is not the only task that JVM performs. The next section looks at
the JVM architecture, because a good understanding of the JVM components can help you in
tuning, maintaining, and monitoring a JRE.

4 IBM Technology for Java Virtual Machine in IBM i5/OS

1.2.1 JVM architecture
Figure 1-3 demonstrates the typical components of JVM.

JVM

JVM API

Interpreter Class loader Memory Diagnostic JITC

management
(GC)

Operating system interface

Figure 1-3 JVM architecture

Interpreter
Interpreter is a computer program that processes Java bytecodes by calling a fixed set of
native instructions for each kind of Java bytecode. This is the slowest way to process Java
bytecodes.

Class loader
Before Java interpreter can start converting Java bytecode, the Java class file has to be
loaded. Java class loader is responsible for supporting Java’s dynamic code loading facilities,
which include the following:
򐂰 Reading standard Java .class files.
򐂰 Resolving class definitions in the context of the current runtime environment. Java
language allows you to define more than one class with the same name. However, these
classes have to be in the different packages2. Class loader is responsible for the correct
class name resolution.
򐂰 Verifying the bytecodes defined by the class file to determine whether the bytecodes are
language-legal.
򐂰 Initializing the class definition after it is accepted into the managed runtime environment.
Class initialization includes memory allocation for the class and assignment of the initial
values to all class parameters and variables.
򐂰 Supporting various reflection APIs for introspection on the class and its defined members.

2
You can have multiple class loaders, and two versions of the same class in the same package can exist in the same
JVM as long as they are loaded with two different class loaders.

Chapter 1. The world of Java 5

 Memory management
The memory management subcomponent is responsible for the efficient use of system
memory by a Java application. Java programs run in a managed execution environment.
When a Java program requires storage, the memory management subcomponent allocates a
discrete region of unused memory called Java heap (or just heap). During runtime a Java
application creates objects in the Java heap. After the application no longer refers to an object
(the storage on the heap), the memory management subcomponent must recognize that the
storage is unused and reclaim the memory for subsequent reuse by the application or return it
to the operating system. This function of the memory management component is called
Garbage Collection (GC).

For further information about memory management subcomponent, refer to Chapter 5,

“Tuning the garbage collector” on page 35.

Diagnostics component
The diagnostics component provides Reliability, Availability, and Serviceability (RAS)
facilities to the JVM. The IBM Virtual Machine for Java is distinguished by its extensive RAS
capabilities. It is designed to be deployed in business-critical operations and includes several
trace and debug utilities to assist with problem determination. If a problem occurs in the field,
it is possible to use the capabilities of the diagnostics component to trace the runtime function
of the JVM and help to identify the cause of the problem. The diagnostics component can
produce output selectively from various parts of the JVM and the just-in-time (JIT) compiler.

JVM API
The JVM API encapsulates all the interaction between external programs and the JVM.
Examples include the following:
򐂰 Creation and initialization of the JVM through the invocation APIs. Creating a JVM involves
native code. JVM API provides Java Native Interface (JNI) API to communicate with the
native code.
򐂰 Handling command-line directives. Any command line arguments that you supply during
JVM startup are processed by JVM API component.
򐂰 Presentation of public JVM APIs such as JNI and JVM Tool Interface (JVMTI). JVMTI
supports monitoring and debugging interfaces for the JVM tools.
򐂰 Presentation and implementation of private JVM APIs used by core Java classes.

Operating system interface

This interface provides the abstraction for the other components of JVM from the underlying
platform. It allows other JVM components to be written in platform neutral manner. Therefore,
you have to replace only operating system interfaces when a JVM vendor, such as IBM,
creates JVM for different platforms.

Just-in-time compiler
Java bytecode interpretation is done by the JVM. It interprets one bytecode at a time. If your
Java program calls the same method twice, the JVM is going to interpret that method two
times. This is just one example. If you look at most of the business programs, string
manipulations, math calculations, and database access methods are among the most typical
operations in such applications.

However, there is a better way to handle interpretation processes of the most frequently used
methods. This is the optimization mechanism that improves the overall performance of Java
applications. This mechanism is implemented in the additional component called just-in-time
compiler.

6 IBM Technology for Java Virtual Machine in IBM i5/OS

 In simple terms the JIT compiler mechanism works as follows:
1. JIT compiler compiles the entire method before it is executed on the system.
2. JIT compiler saves the compiled version of the method on a heap, called JIT heap.
3. When the same method is called again, JIT compiler checks its heap and if the method is
there, it runs without recompilation.
Due to the physical limitations of the main memory, not all compiled methods can be
stored in the JIT heap.

Note: JIT compiler is not a required component of JVM, but it is found in all production
implementations of JVM.

JIT compiler optimization in IBM Technology for Java Virtual Machine

If you try to analyze the usage pattern for all methods executed in JVM, you must see that
some methods are executed a few times and some are executed tens of thousands of times
over the JVM lifetime. It will be nice to keep only the most frequently-used methods on the JIT
heap. If we could implement this technique, JIT compiler would greatly improve the chances
of finding the compiled version of the method on the heap. But how does JIT compiler know
what are the most frequently-used methods?

The following algorithm implements this optimization:

1. When JVM starts, it takes all methods through the interpreter, not JIT compiler.
2. At the same time, JVM counts how many times each method has been interpreted.
3. When this count reaches a predefined number N, it is called JIT threshold. JIT compiler
compiles this method and places it on the JIT heap. The count of this method is reset.
4. Subsequent calls to the same method continue to increment its count. When the call count
of a method reaches a JIT recompilation threshold, the JIT compiles it a second time, this
time applying a larger selection of optimizations than on the previous compilation,
because the method has proven to be a significant part of the whole program. The
recompilation process is iterative. The call count of a recompiled method is reset again
and, as it reaches succeeding thresholds, it triggers recompilation at increasing
optimization levels. Thus, the busiest methods of a Java program are always optimized
most aggressively, maximizing the performance benefits of using the JIT.
5. This process continues during the lifetime of the JVM.

1.3 Java Runtime Environment

Java Runtime Environment provides all necessary components to run Java applications on a
specific platform. In i5/OS you get JRE as part of Java SDK.

Figure 1-1 on page 3 shows the main components of JRE:

򐂰 JVM
This has been described 1.2, “Java Virtual Machine” on page 4.
򐂰 Core classes
As described in 1.1.2, “Java platform and Java software development kit” on page 3, Java
platform has three editions. Each edition includes a predefine set of APIs. Core classes
are the compiled classes for all APIs that are supported by a specific edition of Java
platform. They usually come in a form of Java archive (JAR) files.

Chapter 1. The world of Java 7

 򐂰 Supporting files
This includes multiple property files and additional tools. For example, there are two
executable programs that allow you to support Remote Method Invocation (RMI) of Java
methods. On the other hand, there is the Java Control Panel tool that provides the method
to view and update some of the global parameters related to JRE.

Note: You get JRE in i5/OS by installing 5722JV1 option 8 (J2SE 5.0 32-bit). JRE is
located in the following directory in the Integrated File System (IFS):
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/

Unlike other platforms, there is no separate install for JRE on i5/OS.

1.4 Java platform development tools and API

Java SDK includes JRE and development tools and APIs. The following tools are the most
important tools in SDK:
򐂰 javac: Java compiler. It produces the Java bytecodes.
򐂰 jar : Java archive. You can archive multiple files and packages in a single JAR file for ease
of use and maintenance. You can package your entire application in a JAR file.
򐂰 javadoc: The tool that creates the Java documentation based on the comments in the
source code.
򐂰 jdb: This is a Java debugger.
򐂰 idlj: This is the Interface Definition Language (IDL) to Java compiler. It generates the Java
bindings based on an IDL file. It is used to allow Java programs to participate in Common
Object Request Broker Architecture (CORBA).

In addition to the tools, Java SDK includes examples of Java programs and development
APIs, such as internationalization and IDL API.

8 IBM Technology for Java Virtual Machine in IBM i5/OS

 2

Chapter 2. Exploring IBM Technology for

Java Virtual Machine
This chapter discusses how Java Virtual Machine (JVM) technology has evolved on the IBM
System i platform. This chapter includes the following:
򐂰 2.1, “A history of Classic JVM and IBM Technology for JVM on System i platform” on
page 10
򐂰 2.2, “Comparing IBM Technology for JVM and Classic JVM” on page 13
򐂰 2.3, “Choosing IBM Technology for JVM or Classic JVM” on page 15
򐂰 2.4, “Using IBM Technology for JVM with existing applications” on page 16
򐂰 2.5, “Future JVM developments on i5/OS” on page 16

© Copyright IBM Corp. 2007. All rights reserved. 9

2.1 A history of Classic JVM and IBM Technology for JVM on
System i platform
The Java platform has been available on System i platform since i5/OS (OS/400 at the time)
V4R2, released in 1998. Prior to direct Java support in i5/OS V4R2, IBM delivered the IBM
Toolbox for Java. The IBM Toolbox for Java is a library of classes that enable Java
applications to directly access System i resources such as database files, data queues, data
areas, and callable program objects. The IBM Toolbox for Java provided the conduit for a
whole new set of client/server and host-based business applications to utilize System i
resources.

IBM also released the IBM Toolbox for Java to the open source community as JTOpen. This
enabled Java application developers to create applications that access System i resources,
even if they did not have a System i platform in their IT infrastructure. IBM continues to
enhance both the licensed program product version (5722-JC1) and the JTOpen versions of
IBM Toolbox for Java.

At that time one of the most common usages of Java with System i applications was Java
applets. A Java applet is a Java application that typically runs within the JVM support
included with a Web browser. A Java developer would create an applet that, for example,
displayed a bar graph of data. A Web developer would code a Hypertext Markup Language
(HTML) page to launch this applet within the user’s browser session.

More importantly from a System i perspective, Java applications could run directly on i5/OS.
Application developers could take advantage of the “write once run anywhere” capabilities
that Java afforded and make them available on multiple platforms, including System i
platform. Workstation-based applications that used the IBM Toolbox for Java classes to
access System i resources could run directly on System i platform.

2.1.1 The System i Classic JVM

IBM System i developers were tasked with designing and developing a Java platform for
i5/OS that would not only just run Java bytecode classes according to Sun specification, but
also take advantage of the ease of use, scalability, reliability, and security that the System i
clients had enjoyed for many years. Another key design requirement was optimizing
performance. At the time, Java applications were interpreted and therefore had a tendency to
consume more processor and memory resources than compiled code.

System i platform uses a technology-independent machine interface (TIMI) to separate the

i5/OS operating system and applications from the underlying hardware implementation. IBM
developed a set of new instructions below the machine interface specifically to support Java.
The actual Java code and application programming interface (API) were (and continue to be)
implemented above the machine interface (licensed program product 5722-JV1 on i5/OS
releases) as shown in Figure 2-1.

10 IBM Technology for Java Virtual Machine in IBM i5/OS

See Figure 2-1.

Compilers User Java

Applications Work JDK
Mgmt

Save/Restore
TIMI Instructions
ADDN
JAVA*

CRTPG

TIMI
Database Comm
JVM
Stacks

Trusted Code
Generator Task Security
Mgmt

Figure 2-1 Classic Java platform implementation on System i platform

i5/OS provided interfaces to the Java platform via the Qshell environment. i5/OS also
provided a RUNJVA control language (CL) command that enabled Java applications to be
called directly from a command line or CL program. i5/OS also included a CRTJVAPGM CL
command to optimize application performance and create a reusable program object that
could be saved between system initial program loads (IPLs). See Figure 2-2. This is known as
the direct executable (DE) environment.

Create Java Program (CRTJVAPGM)

Type choices, press Enter.

Class file or JAR file . . . . . > '/home/myfolder/Hello.class'

Classpath . . . . . . . . . . . > '.:/home/myfolder'

Java developer kit version . . . *NONE Character value, *NONE

Optimization . . . . . . . . . . > 30 10, *INTERPRET, 20, 30, 40
User profile . . . . . . . . . . *USER *USER, *OWNER
Use adopted authority . . . . . *NO *NO, *YES
Replace program . . . . . . . . *YES *YES, *NO
Enable performance collection . *NONE *NONE, *ENTRYEXIT, *FULL
Profiling data . . . . . . . . . *NOCOL *NOCOL, *COL

Additional Parameters

Directory subtree . . . . . . . *NONE *NONE, *ALL

Target release . . . . . . . . . *CURRENT *CURRENT,V4R4M0, ...
More...
Figure 2-2 i5/OS CRTJVAPGM code screen

Chapter 2. Exploring IBM Technology for Java Virtual Machine 11

 Java on i5/OS was able to take advantage of the System i 64-bit architecture to provide a
scalable solution from single processor machines all the way up to multiprocessor machines.
The Classic JVM was unique in its implementation of asynchronous garbage collection, which
allowed the JVM to continue processing application requests during the garbage collection
cycle. As Java usage requirements evolved from primarily stand-alone applications to
distributed, Web based applications, IBM has evolved the Classic Java platform environment
as well. WebSphere Application Server is an excellent example. Very early versions of
WebSphere Application Server made extensive use of directly executable programs on
System i platform, because this typically provided better performance than interpreted code.

Developments in JVM technology and just-in-time (JIT) compilers continued to be added to

the Classic JVM, especially in conjunction with IBM WebSphere Application Server. In i5/OS
V5R2, IBM introduced a mixed mode interpretation (MMI) environment where each Java
method has been interpreted a finite number of times (2000 by default). After this
interpretation threshold has been reached, the method is compiled into native code by the JIT
compiler. In recent releases, and especially for WebSphere Application Server based
workloads, JIT compiled methods have tended to perform better than the DE environment.

The Classic JVM is used by default for Java applications run from the i5/OS Qshell
environment, and the RUNJVA CL command is also used. The Classic JVM is also the
default for WebSphere Application Server workloads. IBM releases new versions of the Java
Development Kit (JDK™) by creating new options for the base IBM Developer Kit for Java
licensed program product. Fixes are delivered by the standard System i program temporary
fix (PTF) process. IBM provides a Java group PTF that includes fixes for the specific i5/OS
release (which is V5R4M0) and the software development kit (SDK) versions (which are 1.3,
1.4, and 5.0). This greatly simplifies system management tasks, especially compared with
other platforms that require a separate fix pack installation for each JDK version.

2.1.2 IBM Technology for Java Virtual Machine

1.1.2, “Java platform and Java software development kit” on page 3 discusses the different
Java editions: ME (for small handheld or micro devices), SE (for standard workstation and
server applications), and EE (for distributed and transactional applications). The System i
Classic JVM is based upon the standard edition, similar to the SDKs for other platforms and
operating systems. Creating different JVM implementations for different Java editions and
operating systems is expensive and time consuming. It can also make it difficult to leverage
improvements across the different implementations.

IBM has invested very heavily in Java technologies for over 10 years. IBM research has
recently developed a highly configurable core JVM that you can use for micro edition
implementations, where a small footprint is of utmost importance, and you can also use it for
server and enterprise implementations, where scalability and extensibility are necessary. This
JVM also has configurable garbage collection policies, which are discussed in further detail in
5.2.1, “Garbage collection policies” on page 37, and advanced performance profiling and
optimization features. This new JVM implementation is referred to as IBM Technology for JVM
throughout the rest of this book.

IBM Technology for JVM is available and supported on most of the popular operating
systems, including i5/OS. IBM Technology for JVM is compliant with JSE version 1.4.2 and
version 5.0 specifications. IBM Technology for JVM is available on i5/OS V5R4M0 and later
releases as 5722-JV1, option 8, and is compliant with JSE version 5.

12 IBM Technology for Java Virtual Machine in IBM i5/OS

2.2 Comparing IBM Technology for JVM and Classic JVM
This section discusses about the similarities and differences between IBM Technology for
JVM and Classic JVM. This provides the foundation to help you determine if the IBM
Technology for JVM option is suitable for your System i environment, and if it is more
advantageous than Classic JVM.

2.2.1 Similarities
Before discussing the differences between Classic JVM and IBM Technology for JVM for
System i platform, it is helpful to understand the similarities. The most important similarity is
that both VMs are fully compliant with the JSE specifications, and Java programs can run in
either VM without modification or recompilation. You can use both environments within the
i5/OS Qshell environment to compile (the javac command) and run (the java command) Java
programs. Both are shipped as part of the 5722-JV1 (IBM Developer Kit for Java).

Both can use the RUNJVA CL command, although some parameters for IBM Technology for
JVM are ignored. Both can use environment variables defined by the i5/OS ADDENVVAR CL
command. Both have their fixes bundled with the i5/OS Java group PTF (SF99291 for
V5R4M0). Both can use the IBM Toolbox for Java to access resources such as data queues
and data areas. Both can use the System i native Java Database Connectivity (JDBC™)
driver to access database resources. Both can use System i debugging API’s that are started
via STRSRVJOB or STRDBG CL commands.

2.2.2 Differences
There are differences between IBM Technology for JVM and Classic JVM. Some are based
upon how IBM Technology for JVM is implemented on System i platform. Others are based
on the different code bases used between the two.

System i platform implementation differences

The Classic JVM is implemented primarily below the System i TIMI and is available on all
currently supported i5/OS releases. The IBM Technology for JVM is implemented in the i5/OS
portable application solutions environment (PASE) and is currently available on i5/OS
V5R4M0, as shown in Figure 2-3.

PASE is discussed in more detail in 3.1, “i5/OS portable application solutions environment” on
page 20. The PASE on System i platform does not use an emulation or operating system
layer and tends to be very fast. The PASE can access System i resources such as the file
system, programs, and Transmission Control Protocol/Internet Protocol (TCP/IP) sockets.
Therefore IBM Technology for JVM can access the same resources as Classic JVM.

Chapter 2. Exploring IBM Technology for Java Virtual Machine 13

 See Figure 2-3.

PASE (AIX)
Applications and
shared libraries

IBM SDK
Compilers User Java
Applications Work
Mgmt JDK 32-bit JVM

Save/Restore
ADDN TIMI Instructions
syscall
JAVA*
CRTPG TIMI

Comm
Database JVM
Stacks
Trusted Code

syscall
Task
Generator Security
Mgmt

PowerPC
PowerPC

Figure 2-3 IBM Technology for JVM implementation on System i platform

The PASE on System i platform supports 32-bit and 64-bit address spaces. With i5/OS
V5R4M0, IBM Technology for JVM currently uses a 32-bit address space. This limits the
theoretical maximum memory usage of IBM Technology for JVM to approximately 4 GB (232
bytes) versus approximately 17 billion GB (264 bytes) for Classic JVM. The IBM Technology
for JVM uses approximately 20% of the memory for its own internal operations, limiting the
maximum heap to approximately 3.25 GB (the practical limit for most applications is in the 2.5
GB to 3.0 GB range). Classic JVM has a maximum heap of 243 GB.

Even though the IBM Technology for JVM’s use of a 32-bit address space limits the maximum
heap size compared with Classic JVM, this does have advantages. The 32-bit address space
uses less memory for the JVM, compared with Classic JVM. As a rule of thumb, the heap size
required for the IBM Technology for JVM is approximately 60% of the Classic JVM, all things
being equal. Suppose you are running WebSphere Application Server workloads today with
the Classic JVM and have a heap size of 4 GB. That same environment running with IBM
Technology for JVM would be approximately 2.5 GB and most likely be able to run fine.

The IBM Technology for JVM is displayed as an application in the PASE. This environment is
very similar to the implementation of IBM Technology for JVM in AIX®. It yields good
performance because it is so close to the processor and has a relatively short code path.
Because it is not implemented above the System i TIMI, it does not have implementation
independence from the underlying hardware changes. However, this must only impact the
JVM and not end user applications.

The i5/OS CRTJVAPGM and DSPJVAPGM commands are not applicable with the IBM
Technology for JVM. Also, the RUNJVA command parameters such as OPTIMIZE are
ignored.

14 IBM Technology for Java Virtual Machine in IBM i5/OS

 Other differences
One of the biggest differences is in garbage collection algorithms available in IBM Technology
for JVM. The Classic JVM uses a System i platform unique algorithm, and typically the only
tuning parameter is the initial and maximum heap size (although most customers set the
maximum heap size parameter to *NOMAX). The IBM Technology for JVM has several
garbage collection algorithms and gives the option of setting the initial heap size and
maximum heap size to an appropriate value. Many other options are available for fine-tuning
GC behavior, which is a further differentiator compared to Classic JVM. Chapter 4, “Making
the switch to IBM Technology for JVM” on page 27 covers this in more detail.

Another difference is that many applications must see performance improvements such as
decreased response time and higher throughput. One reason is because of the smaller
memory requirements: object references are 32 bits with IBM Technology for JVM, compared
with 64 bits for Classic JVM. Also, IBM has optimized the IBM Technology for JVM for the
PASE. The calling methods for applications running in IBM Technology for JVM that utilize
System i services (such as Integrated Language Environment® (ILE) programs and TCP/IP
sockets) incur a slight increase in overhead compared to Classic JVM. However, the
execution methods within applications running in IBM Technology for JVM incur less
overhead than with Classic JVM and typically more than compensate for the calling methods.

Finally, there are substantial differences in monitoring agents and diagnostic tools. Both VMs
support a set of tools. However, in most cases these tools are different for each VM. For
System i developers, who have been working with the native tools, Table 2-1 shows the
support of native tools for IBM Technology for JVM.

Table 2-1 JVM performance tools compatibility

Tool Usage with IBM Technology for JVM

ANZJVM command No

DMPJVM command No

Service tools JavaGCInfo macro No

iDoctor for IBM iSeries™ Heap Analysis Tools for Java No

Performance Explorer Some

Tivoli® Performance Viewer (WebSphere Application Yes

Server)

The new set of tools for IBM Technology for JVM is described in Chapter 8, “Analyzing JVM
behavior” on page 83.

There are a few other differences that are listed in 2.4, “Using IBM Technology for JVM with
existing applications” on page 16.

2.3 Choosing IBM Technology for JVM or Classic JVM

Choosing the appropriate Java version 5 SDK must not be a difficult exercise. System i
platform gives you the flexibility to use IBM Technology for JVM and Classic JVM. You can try
both for your specific applications and System i environment, and choose the better
performer, considering lower response time, more throughput, and so on.

Chapter 2. Exploring IBM Technology for Java Virtual Machine 15

 However, if you are using i5/OS V5R3 you do not have the option to use IBM Technology for
JVM. WebSphere Application Server version 6.0 and earlier releases do not use Java version
5, therefore the Classic JVM is your only option for pre-WebSphere Application Server 6.1
workloads, even if you are using i5/OS V5R4.

If you do have i5/OS V5R4 or later and, optionally, WebSphere Application Server version 6.1
or later, then how do you choose? If scalability in a single JVM process on a large
multiprocessor System i platform (which means eight or more processors) or i5/OS
integration are important, then Classic JVM might be a better fit.

In general, you must estimate the memory requirements for your application running in IBM
Technology for JVM. If the application’s memory requirements fit in 32-bit addressable space,
you must test your application in both VMs to choose the best performing one. Refer to
Chapter 4, “Making the switch to IBM Technology for JVM” on page 27 for more details.

In our tests IBM Technology for JVM, on average, has been 7-10% faster than Classic JVM.

2.4 Using IBM Technology for JVM with existing applications

Generally speaking, applications that run within Classic JVM (5722-JV1, option 7) must run
within IBM Technology for JVM (5722-JV1, option 8). For example, WebSphere Application
Server version 6.1 is the first version that runs on the Java version 5 SDK. This is supported
on Classic JVM and IBM Technology for JVM, on i5/OS V5R4M0. There are a few caveats
you must be aware of:
򐂰 Applications running within IBM Technology for JVM do not support i5/OS adopted
authority. Refer to 7.3.4, “Adopted authority” on page 77.
򐂰 ILE C programs that are linked must be compiled with the teraspace option on (STGMDL
*TERASPACE), if used with IBM Technology for JVM.
򐂰 PASE 64-bit address space based applications are not supported on IBM Technology for
JVM.

2.5 Future JVM developments on i5/OS

IBM has, and continues to make, substantial investments in Java technology. At the same
time, IBM also understands your requirements to preserve your investment in applications
and skills, and to implement new technologies according to your own timetable. Accordingly,
the Classic JVM will still be supported in the next i5/OS release. This gives you the option of
evaluating and testing your applications on IBM Technology for JVM, at a pace suitable for
your organization.

IBM Technology for JVM represents IBM strategic JVM offering. Using a common code base
across multiple platforms provides a focus for IBM research and development investment in
Java technology.

In the future you can expect to see a 64-bit version of the IBM Technology for JVM. This
enables you to compare the 32-bit and 64-bit versions of the IBM Technology for JVM to
determine which is optimal for your specific environment. The eventual introduction of 64-bit
IBM Technology for JVM on i5/OS brings i5/OS in line with other platforms such as AIX and
Linux®, for which both 32-bit and 64-bit versions of IBM Technology for JVM are available.

16 IBM Technology for Java Virtual Machine in IBM i5/OS

You can expect to see performance improvements, and also improved monitoring and
analysis tools as the IBM Technology for JVM continues to mature. Some of these tools are
available on multiple platforms. Some may be System i platform unique.

Finally, you can expect to see application development tools such as WebSphere
Development Studio Client for iSeries, incorporate IBM Technology for JVM. This will enable
developers to extensively test and debug their stand-alone applications and WebSphere
Application Server based IBM Technology for JVM applications prior to putting them into
production on a System i platform.

Chapter 2. Exploring IBM Technology for Java Virtual Machine 17

18 IBM Technology for Java Virtual Machine in IBM i5/OS
 3

Chapter 3. New user guide

This chapter discusses how you can start using the IBM Technology for Java Virtual Machine
(JVM). This chapter covers the following topics:
򐂰 3.1, “i5/OS portable application solutions environment” on page 20
򐂰 3.2, “Installing IBM Technology for JVM” on page 20
򐂰 3.3, “Basic configuration” on page 23
򐂰 3.4, “Known issues” on page 25

© Copyright IBM Corp. 2007. All rights reserved. 19

3.1 i5/OS portable application solutions environment
As mentioned in Chapter 1, “The world of Java” on page 1 and Chapter 2, “Exploring IBM
Technology for Java Virtual Machine” on page 9, the IBM Technology for Java Virtual Machine
is targeted to be the IBM main virtual machine for Java, based on a common, high-speed,
modular code base.

To maximize the reuse of common code, i5/OS uses the AIX version of the IBM Technology
for JVM as its base. The code is customized for use under the i5/OS portable applications
solutions environment (i5/OS PASE). Because most of the code is common, the i5/OS
version is able to rapidly adapt updates and fixes made in IBM Technology for JVM,
especially those updates with AIX-specific benefits or considerations.

3.2 Installing IBM Technology for JVM

A successful install involves several steps. This section discusses the necessary
prerequisites, the installation, and verification.

3.2.1 Checking prerequisites

To ensure a successful installation of the IBM Technology for JVM you must make sure that
your system meets the following prerequisites:
򐂰 Make sure that your system is running i5/OS V5R4. If your system is running a version of
i5/OS that is older than V5R4 you have to upgrade before you can use IBM Technology
for JVM.
򐂰 Your system must also have i5/OS PASE installed. To verify that i5/OS PASE is installed
perform the following steps:
a. Enter GO LICPGM command.
b. Select 10 (Displays installed licensed programs).
c. Press F11 until you see the column product option open.
d. Page down and look for option 33 (Portable App Solutions Environment) of 5722SS1
as shown in Figure 3-1.
If you find that i5/OS PASE is not installed, review the iSeries Information Center, V5R4
for installation instructions on the Web at:
http://publib.boulder.ibm.com/infocenter/iseries/v5r4/topic/rzalf/rzalfgetstart
.htm
򐂰 Ensure that the 5722JV1 product with option *BASE is installed on your system

It is also important to ensure that the latest program temporary fixes (PTFs) are installed.
i5/OS requires:
򐂰 Latest cumulative PTF package
򐂰 Latest HIPER Group PTF (SF99539)

Note: For more information about required PTFs, visit the following Web site:
http://www.ibm.com/systems/support/i/fixes/index.html

20 IBM Technology for Java Virtual Machine in IBM i5/OS

 See Figure 3-1.

Display Installed Licensed Programs

System: RCHAS60
Licensed Product
Program Option Description
5722SS1 18 Media and Storage Extensions
5722SS1 21 Extended NLS Support
5722SS1 22 ObjectConnect
5722SS1 23 OptiConnect
5722SS1 25 NetWare Enhanced Integration
5722SS1 26 DB2 Symmetric Multiprocessing
5722SS1 27 DB2 Multisystem
5722SS1 29 Integrated Server Support
5722SS1 30 Qshell
5722SS1 31 Domain Name System
5722SS1 33 Portable App Solutions Environment
5722SS1 34 Digital Certificate Manager
5722SS1 35 CCA Cryptographic Service Provider
5722SS1 36 PSF 1-45 IPM Printer Support
More...
Press Enter to continue.

F3=Exit F11=Display status F12=Cancel F19=Display trademarks

Figure 3-1 Verifying that i5/OS PASE is installed

For i5/OS PASE, the list of required PTFs can be found here:
http://www-03.ibm.com/servers/enable/site/porting/iseries/pase/misc.html

3.2.2 Installing IBM Technology for JVM

Installing the IBM Technology for JVM product is a simple process.

IBM Technology for JVM is included in licensed program 5722-JV1 (option 8). The install
media for 5722-JV1 is included on the V5R4 Standard Set Media. Option 8 of 5722-JV1 is
typically found on the CD labeled D29xx_07.

The following steps install IBM Technology for JVM:

1. Make sure the install CD is in the System i optical drive.
2. Start the installation with the following command:
RSTLICPGM LICPGM(5722JV1) DEV(OPT01) OPTION(8)
This is assuming OPT01 is the name of your optical device.

When the installation completes you have to install the latest Java group PTF (SF99291). An
easy way to install SF99291 is by using the GO PTF menu and take option 8 (install program
temporary fix package). When this book was written, SF99291 was at level 4. You can use
the following Web site to check if there is a more current version:
http://www.ibm.com/systems/support/i/fixes/index.html

On a CL command line you can use the WRKPTFGRP command to verify the level of each
group PTF installed on the system.

Chapter 3. New user guide 21

3.2.3 Verifying the installation
With the PTFs installed, the final step is to verify that the new product is functional. Run the
following commands:
1. Open a Qshell session:
STRQSH
2. Add JAVA_HOME to your shell environment:
export JAVA_HOME=/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit
3. Invoke the Hello program included with the Java product (5722JV1) to verify the
installation:
java -showversion -classpath /qibm/proddata/java400 Hello
Figure 3-2 shows the example of running this command. Notice that the dates for the JVM
component builds may be different when you run this program on your system.

Note: By adding -showversion to the java command additional lines of output are
displayed that indicate which JVM is in use.

QSH Command Entry

$
export JAVA_HOME=/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit
$
java -showversion -classpath /qibm/proddata/java400 Hello
java version "1.5.0"
Java(TM) 2 Runtime Environment, Standard Edition (build jclap32dev)
IBM J9 VM (build 2.3, J2RE 1.5.0 IBM J9 2.3 OS400 ppc-32 (JIT enabled)
J9VM - 20060501_06428_bHdSMR
JIT - 20060428_1800_r8
GC - 20060501_AA)
JCL - jclap32dev

Hello World
$
Figure 3-2 Running the Hello program to verify IBM Technology for JVM installation

If your system has a java.version=1.3 or 1.4 specified in a property file, you may still see an
output for Classic JVM. Adding the property -Djava.version=1.5 on a command line assures
that you actually use IBM Technology for JVM, for example:
java -showversion -Djava.version=1.5 -classpath /qibm/proddata/java400 Hello

3.2.4 Uninstalling IBM Technology for JVM

If you have to remove the IBM Technology for JVM for any reason, you can use the following
command:
DLTLICPGM LICPGM(5722JV1) OPTION(8)

22 IBM Technology for Java Virtual Machine in IBM i5/OS

3.3 Basic configuration
This section discusses how to set up your environment to use IBM Technology for JVM. If
both Classic JVM and IBM Technology for JVM are installed, the default behavior on i5/OS
V5R4 is to use Classic JVM.

3.3.1 Setting JAVA_HOME

As mentioned in 3.2.3, “Verifying the installation” on page 22 we set JAVA_HOME before
running the Hello Java program. Setting JAVA_HOME to the installation directory of IBM
Technology for JVM toggles between Classic JVM and IBM Technology for JVM.

In order to activate the IBM Technology for JVM, JAVA_HOME must be set to the following:
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit

Note: This value is case sensitive. If JAVA_HOME is set to an invalid directory, any attempt
to invoke Java fails with an error similar to Figure 3-3.

If you are familiar with the Classic JVM on i5/OS you know that you can have multiple JDKs
installed (1.3, 1.4, 1.5, and so on) and you can switch between them by setting the
java.version property. The only way to use IBM Technology for JVM is by setting the
JAVA_HOME environment variable. The behavior of JAVA_HOME is as follows:
򐂰 If JAVA_HOME is not set, or set to the empty string, all forms of Java invocation uses the
i5/OS default Classic JVM implementation. If the Classic JVM is not installed, IBM
Technology for JVM is used.
򐂰 If java.version is set to 1.3 or 1.4, JAVA_HOME is ignored and Classic JVM is used.
򐂰 If JAVA_HOME is set to a valid JVM installation directory, all Java invocations use the
specified VM.
򐂰 If JAVA_HOME is set to any other value, Java invocation fails with an error message as
shown in Figure 3-3.

Java Shell Display

JAVA_HOME directory /BLAH not found. Java Virtual Machine not created.
Java program completed with exit code 1
Figure 3-3 Showing the error message displayed when JAVA_HOME is not set correctly

Three techniques you can use to set JAVA_HOME

Perform the following steps:
1. Use the ADDENVVAR CL command. For example, to add the variable to your current
interactive job enter the following on a CL command line:
ADDENVVAR ENVVAR(JAVA_HOME) VALUE('/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit’)
Use the following if you want to set the variable at the system level:
ADDENVVAR ENVVAR(JAVA_HOME) VALUE('/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32
bit') LEVEL(*SYS)

Chapter 3. New user guide 23

 Note: Use caution when setting JAVA_HOME to level *SYS. All subsequent Java
invocations can be affected if they do not explicitly set JAVA_HOME in their respective
startup routines.

If you want to remove the JAVA_HOME variable type, use the following:
RMVENVVAR JAVA_HOME
2. Set JAVA_HOME from within QSHELL. For example:
a. Open a Qshell session using:
STRQSH
b. Export the variable to your process:
export -s JAVA_HOME=/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit
To remove the JAVA_HOME environment variable from within Qshell use the unset
command. For example, use the following:
unset JAVA_HOME
3. Use a .profile file to initialize the shell. If you would like JAVA_HOME to be set every time
you start Qshell, you can create a .profile file in a user’s home directory. For example, if
your i5/OS user profile is John you would perform the following steps:
a. Create a file called .profile in /home/John.
b. Add the following text to the file:
export -s JAVA_HOME=/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit
c. Save the file.
Now, when you enter Qshell it initializes the shell with JAVA_HOME. From within the
Qshell session, run the env command to confirm that JAVA_HOME is set. See the sample
output in Figure 3-4.

QSH Command Entry

$
env
LANG=/QSYS.LIB/EN_US.LOCALE
JAVA_HOME=/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit
QIBM_USE_DESCRIPTOR_STDIO=I
TRACEOPT=UNLINK
QIBM_DESCRIPTOR_STDERR=CRLN=N
QIBM_DESCRIPTOR_STDOUT=CRLN=N
QIBM_DESCRIPTOR_STDIN=CRLN=Y
LOGNAME=PAULS
SHLVL=1
HOSTTYPE=powerpc
HOSTID=123.123.123.123
HOSTNAME=RCHAS60.SERVER.IBM.COM
OSTYPE=os400
MACHTYPE=powerpc-ibm-os400
TERMINAL_TYPE=5250
...
Figure 3-4 JAVA_HOME is set correctly

24 IBM Technology for Java Virtual Machine in IBM i5/OS

3.3.2 Setting default command line options in SystemDefault.properties file
If you are familiar with using the Classic JVM, you have probably used the
SystemDefault.properties file to set a Java system property or command-line option. It is an
i5/OS specific mechanism and it also works for IBM Technology for JVM.

If you prefer to use a method that works on all platforms, you can refer to Appendix D of the
IBM Developer Kit and Runtime Environment, Java 2 Technology Edition, Version 5.0
Diagnostics Guide, available at:
http://download.boulder.ibm.com/ibmdl/pub/software/dw/jdk/diagnosis/diag50.pdf

You can create the SystemDefault.properties file in either of the following locations:
򐂰 Your user.home directory:
For example: /home/John/SystemDefault.properties
򐂰 The /QIBM/UserData/Java400 directory:
/QIBM/UserData/Java400/SystemDefault.properties

If you prefer to customize the name and location of the properties file you can define the
QIBM_JAVA_PROPERTIES_FILE environment variable. The following is a sample of a CL
command:
ADDENVVAR ENVVAR(QIBM_JAVA_PROPERTIES_FILE)
VALUE(/qibm/userdata/java400/mySystem.properties)

It is certainly possible to have a SystemDefault.properties file in multiple locations. In that

case the order of precedence is as follows:
1. Command line or JNI invocation API
2. QIBM_JAVA_PROPERTIES_FILE environment variable
3. The /home/<userid>/SystemDefault.properties file
4. /QIBM/UserData/Java400/SystemDefault.properties
5. Default system property values

For additional details and examples review Chapter 17 in the IBM Developer Kit and Runtime
Environment, Java 2 Technology Edition, Version 5.0 Diagnostics Guide, available at:
http://download.boulder.ibm.com/ibmdl/pub/software/dw/jdk/diagnosis/diag50.pdf

3.4 Known issues

There are a few known issues with running IBM Technology for JVM in i5/OS. See Table 3-1
for the details.

Table 3-1 Known problems

Problem Cause Possible Solution

java.class.path property and Classpath is being overridden Modify system properties file
$CLASSPATH environment by a java.class.path entry in a accordingly.
variable are ignored. system properties file.

IBM Technology for JVM does not QIBM_JAVA_PASE_START Set

run my PASE native method from UP environment variable is QIBM_JAVA_PASE_START
Qshell. It fails with a not set or is set improperly. UP to '/usr/lib/start32'.
java.lang.UnsatisfiedLinkError.

Chapter 3. New user guide 25

 Problem Cause Possible Solution

Output is unreadable on screen. The os400.stdio.convert Check properties files to make

property may be set to 'N'. sure that os400.stdio.convert
is not altered.

IBM Technology for JVM does not This is not a standard feature Modify your program to
support adopted authority. of a JVM. remove the adopt authority
dependency (refer to 7.3.4,
“Adopted authority” on
page 77).

Call to native program fails. Program was not compiled to Recompile the native program
support teraspace. with the following options:
TERASPACE(*YES)
STGMDL(*TERASPACE)
DTAMDL(*LLP64).

IBM Technology for JVM does not

support calls to the 64-bit based
programs in PASE.

JAVA_HOME points to IBM java.version is set to 1.3 or 1.4 Change java.version to 1.5.
Technology for JVM, but your Java in one of the property files.
program starts in Classic JVM.

26 IBM Technology for Java Virtual Machine in IBM i5/OS

 4

Chapter 4. Making the switch to IBM

Technology for JVM
This chapter provides information that helps you decide if switching to IBM Technology for
JVM is right for your application. This chapter covers the following:
򐂰 4.1, “Fitting your application into a 32-bit JVM” on page 28
򐂰 4.2, “Configuring garbage collection” on page 30
򐂰 4.3, “Finding dependencies to the Classic JVM” on page 32
򐂰 4.4, “WebSphere Application Server and IBM Technology for JVM” on page 33

© Copyright IBM Corp. 2007. All rights reserved. 27

4.1 Fitting your application into a 32-bit JVM
When deciding to switch to the new IBM Technology for JVM you have to consider that you
are moving from a 64-bit to a 32-bit implementation. This means that every reference to an
object is half the size. This does not imply that your application now requires 50% less
memory. Some early testing has shown that 20-50% reduction in heap (memory footprint) is a
reasonable expectation. Therefore, now you must be able to make a quick calculation to see
if your application is a candidate to fit in IBM Technology for JVM. A general rule of thumb is
that any Classic JVM that is less than about 5 GB in size must run in IBM Technology for
JVM. After you switch to the IBM Technology for JVM, be sure to test your application to
ensure that any peaks in workload do not result in an OutOfMemoryError.

Tip: A general rule of thumb is that any application running under Classic JVM with a heap
that is less than about 5 GB in size must run in IBM Technology for JVM without a problem.

The general procedure recommended for switching an application from Classic JVM to IBM
Technology for JVM is as follows:
1. Measure the heap usage of the application under Classic JVM. Refer to 4.1.1, “Measuring
heap usage in Classic JVM”. If the heap usage is significantly larger than 5 GB under
Classic JVM, the application may run out of heap space under IBM Technology for JVM.
If you are trying to run a new application, skip this step.
2. Estimate the required maximum heap size under IBM Technology for JVM based on the
usage under Classic JVM. You can do this if you multiply your heap requirement in Classic
JVM by 0.6.
If you are testing a new application, or are not certain about the performance
characteristics of an existing application running in the Classic 64-bit VM, start by running
the application in IBM Technology for JVM with the default heap size parameters
(currently an initial heap size of 4 MB and a maximum of 2 GB).
3. Determine heap settings based on estimated heap requirements. Refer to 5.2.2, “Tuning
heap size” on page 41 for information about how to change heap settings.
4. Test application under load using IBM Technology for JVM, collecting verbose garbage
collection (GC) data. Keep the default optthruput GC policy, but use the heap settings
from step 3 listed previously.
Refer to 5.2.3, “Verbose GC output” on page 44 for information about how to collect
verbose GC data.
If performance/memory usage do not meet expectations or OutOfMemoryErrors are
generated, then perform the following steps:
a. Analyze heap usage using EVTK or manual review of verbose GC data. Refer to
“Troubleshooting garbage collection” on page 133 for information about using EVTK.
b. Adjust heap settings and repeat test cycle until satisfactory performance is achieved.
5. Optionally, once the application is using a suitable heap size, you may wish to test
different GC policies and how they interact with your application.
Refer to 5.2.1, “Garbage collection policies” on page 37 for more information about the
available GC policies and how to use a different policy.

The remainder of this chapter provides guidance on some of these steps and some other
considerations for switching to IBM Technology for JVM.

28 IBM Technology for Java Virtual Machine in IBM i5/OS

4.1.1 Measuring heap usage in Classic JVM
If you do not really know how much memory your application uses, you have to obtain this
information before you can make an informed decision about switching to IBM Technology for
JVM. There are a couple of techniques you can use to find the memory footprint of the
Classic JVM.

Dump Java Virtual Machine (DMPJVM) command

The DMPJVM command dumps information about the JVM for a specified job. It includes
information about the classpath, garbage collection, and threads associated with the JVM.
The output goes to a spool file for the user that executes the command as shown in
Figure 4-1.

Attention: DMPJVM does hold the job briefly while it takes a snap-shot of the JVM. On
rare occasions, with large JVMs, this can cause a performance problem that could require
a JVM restart to recover.

The following is a sample CL command:

DMPJVM JOB(186947/QEJBSVR/WAS60SVR)

........................................................................
. Garbage Collection .
........................................................................
Garbage collector parameters
Initial size: 98304 K
Max size: 240000000 K
Current values
Heap size: 453768 K
Garbage collections: 115
Additional values
JIT heap size: 188256 K
JVM heap size: 310524 K
Last GC cycle time: 449 ms
........................................................................
Figure 4-1 Showing total heap

Figure 4-1 shows an extract of the DMPJVM output. You have to use Heap size value from the
output to estimate if this application fits into the IBM Technology for JVM.

Note: If the DMPJVM command does not produce any output, the command may have
been terminated by your interactive job. Check the Default wait time of your job using the
CHGJOB command. It has a default value of 30 seconds. You may want to try increasing
this to 300 seconds to give the command sufficient time to complete.

For more information about the DMPJVM command, refer to the System i Information Center,
V5R4 at the following link:
http://publib.boulder.ibm.com/infocenter/iseries/v5r4/index.jsp

Chapter 4. Making the switch to IBM Technology for JVM 29

 Using the JAVAGCINFO advanced analysis command
The JAVAGCINFO command is much more cumbersome to use compared to DMPJVM but it
must not have any negative impact to the JVM. The JAVAGCINFO command requires a
sign-in to Start Service Tools (STRSST). Perform the following steps to gather the heap
statistics:
1. Issue STRSST at a CL command line.
2. Sign in with an appropriate ID.
3. Select 1. Start a service tool.
4. Select 4. Display/Alter/Dump.
5. Select 1. Display/Alter storage.
6. Select 2. Licensed Internal Code (LIC) data.
7. Select 14. Advanced analysis. You may have to page/scroll down to see this option.
8. Find JAVAGCINFO in the list, type 1 in the option field to select, press Enter.
9. On the “Specify Advanced Analysis Options” just press Enter with no options.

What you see on your screen is some basic information about every Classic JVM on the
system that is active. To locate the job you are interested in, look for the task name.
Figure 4-2 illustrates what you will see for a job named “WAS60SVR.”

JVM GENERAL INFORMATION:

VM*=EBE9DA08E4056A00 TASK=B00290000FB29000 TASK NAME=MITHREAD-WAS60SVR QEJBSVR 186947
...
JVM HEAP SIZE INFORMATION. SIZES ARE IN BYTES:
INITIAL HEAP SIZE=100663296 THRESHOLD PER NODE = 100663296 NODES = 8
INITIAL MAXIMUM HEAP SIZE=*NOMAX CURRENT MAXIMUM HEAP SIZE=*NOMAX
GC HEAP MANAGER*=F85E62169E002D80 SIZE=513261568 JVM HEAP*=F85E62169E000048 SIZE=298545152
JVM PROBLEM MODE HEAP*=C77A2BBB34000048 SIZE=20545536 JIT HEAP=C3657B7BA9000048
SIZE=193298432
Figure 4-2 JAVAGCINFO output

The various heap sizes (in bytes) are shown in Figure 4-2. You have to use the first value of
heap (SIZE=513261568) from the output to estimate if this application fits into IBM
Technology for JVM.

4.2 Configuring garbage collection

IBM Technology for JVM offers more options for configuring memory management
component than the Classic JVM. This section describes the most important options.

4.2.1 Choosing the right policy

Chapter 5, “Tuning the garbage collector” on page 35 describes the various garbage
collection policies that are available with IBM Technology for JVM. Garbage collection in
general is completely different compared to the Classic JVM. Typically, GC is something that
users must not have to worry about but there are a couple of things to keep in mind when you
are switching to the IBM Technology for JVM.

30 IBM Technology for Java Virtual Machine in IBM i5/OS

 The most important thing to remember when comparing the GC behavior of the 2 VMs is that
the minimum heap size parameter means something very different to each environment. The
Classic JVM has its own way of doing garbage collection and it does not allow the user to
alter the behavior. The only thing that a user can influence is the frequency of GC cycles. This
is done by changing the initial heap size at VM startup. For example, if you set the initial heap
size to 64 MB, (-Xms64m) GC runs for the first time after 64 MB of space has been allocated.
When a GC cycle is triggered, it frees all the space that is no longer referenced by other
objects in the VM. Subsequent GC cycles run when another 64 MB of space has been
allocated. Therefore, in the case of the Classic JVM, the initial heap size is really not what the
name implies, it is more similar to a GC allocation threshold. This has been a point of
confusion for Java programmers that are not familiar with Java on i5/OS. Users of the new
IBM Technology for JVM are going to find that the initial heap size simply sets the initial heap
to the value specified as its name implies. The behavior of GC is controlled by the GC policy.

Therefore now you might be wondering which GC policy is best for your application. The
recommendation is that you always start with the default GC policy. The default policy,
optimize for throughput (optthruput) is typically used for applications where raw throughput is
more important than short GC pauses. The application is stopped each time that garbage is
collected. If you prefer to test another policy, it is recommended that you read about each
policy in Chapter 5, “Tuning the garbage collector” on page 35.

Tip: The recommendation is that you always start with the default GC policy. The default
policy, optimize for throughput (optthruput) is typically used for applications where raw
throughput is more important than short GC pauses.

4.2.2 The minimum and maximum heap size

In 4.2.1, “Choosing the right policy” on page 30, this chapter discusses how the minimum
heap size (-Xms) is a special GC parameter for the Classic JVM. When you switch to IBM
Technology for JVM the minimum heap size simply defines the minimum size of the VM. This
is a significant change, therefore you must make sure that the minimum and maximum heap
size are set to values that make sense for the new IBM Technology for JVM. Table 4-1 shows
the default values.

Table 4-1 Default heap sizes

VM Default value for -Xms Default value for -Xmx

Classic JVM 16MB *NOMAX

IBM Technology for JVM 4MB 2GB

To give an example we include Table 4-2 to show you how the WebSphere Application Server
product changes these values when switching between Classic JVM and IBM Technology for
JVM. For additional help choosing appropriate values for minimum and maximum heap, refer
to 5.2.2, “Tuning heap size” on page 41.

Table 4-2 Default heap size values for WebSphere Application Server V6.1
VM Default value for -Xms Default value for -Xmx

Classic JVM 96MB *NOMAX

IBM Technology for JVM 50MB 256MB

Chapter 4. Making the switch to IBM Technology for JVM 31

4.3 Finding dependencies to the Classic JVM
There must not be too many dependencies to be concerned about, but this section has a few
tips that might help you have a smoother transition to the IBM Technology for JVM:
򐂰 Check for any system properties that might not be supported. For example:
os400.gc.heap.size.max
This property is an alternative to using -Xmx (setting maximum GC size) for the Classic
JVM, but it does not have any affect on the IBM Technology for JVM. If you are using this
property with the Classic JVM, it is recommended that you use the -Xmx instead. The
same is true for os400.gc.heap.size.min, use the -Xms command line argument.
The i5/OS Information Center has a complete list of supported properties for Classic JVM.
Use the following link to view them:
http://publib.boulder.ibm.com/infocenter/iseries/v5r4/topic/rzaha/sysprop2.htm
Most of the properties found in the i5/OS Information Center are only applicable to Classic
JVM. Table 4-3 shows properties that are applicable to both Classic JVM and IBM
Technology for JVM.

Table 4-3 Common properties of Classic JVM and IBM Technology for JVM
Properties Description

file.encoding Maps the coded character set identifier (CCSID) to the corresponding ISO
American Standard Code for Information Interchange (ASCII) CCSID. Also,
sets the file.encoding value to the Java value that represents the ISO ASCII
CCSID. See file.encoding values and iSeries CCSID for a table that shows the
relationship between possible file.encoding values and the closest matching
CCSID. ISO8859_1 is the default value

java.class.path Designates the path that i5/OS uses to locate classes. Defaults to the
user-specified CLASSPATH.

java.compiler Specifies whether you use the just-in-time (JIT) compiler or not. For Classic
JVM the value can be jitc (JIT compiler) or jitc_de (both the JIT compiler
and direct execution).
For IBM Technology for JVM the values are NONE or j9jit23. If you specify any
other value, JIT compiler just ignores the setting and acts as though you
specified j9jit23.

os400.stdio.convert Allows control of the data conversion for stdin, stdout, and stderr in Java. Data
conversion occurs by default in the Java virtual machine to convert ASCII data
to or from Extended Binary Coded Decimal Interchange Code (EBCDIC). You
can turn these conversions on or off with this property, which affects the
current Java program.

Note: Refer to 4.2, “Configuring garbage collection” on page 30 for recommended

values of maximum and minimum heap size.

򐂰 If your Java program uses adopted authority, it has to be modified. Refer to 7.3.4,
“Adopted authority” on page 77 for more information.

32 IBM Technology for Java Virtual Machine in IBM i5/OS

 򐂰 If your application calls native methods, they must be recompiled for teraspace as
described in 7.3.1, “Teraspace storage model” on page 76.
򐂰 This might seem obvious but the best way to find a dependency is to give the IBM
Technology for JVM a try. If it fails, try the following:
a. Verify that it was working at the same JDK when it was running in Classic JVM mode.
The problem can be JDK related if it was working at JDK 1.4.
b. Look for any environment variables or system properties that can cause a problem.
Use 3.3, “Basic configuration” on page 23 for reference.

4.4 WebSphere Application Server and IBM Technology for

JVM
To answer the question, “Does WebSphere Application Server use IBM Technology for
JVM?.” You can configure WebSphere Application Server V6.1 to use the Classic JVM or the
IBM Technology for JVM. The default setting is to use the Classic JVM.

To make sure that switching WebSphere Application Server to use IBM Technology for JVM
is a good idea, review the previous topics in this chapter:
򐂰 Refer to 4.1, “Fitting your application into a 32-bit JVM” on page 28. Notice, however, that
the practical limit for the Java heap running WebSphere Application Server is 2.5 GB.
򐂰 Refer to 4.3, “Finding dependencies to the Classic JVM” on page 32.

If you decide to switch your WebSphere Application Server to use IBM Technology for JVM,
then refer to Appendix A, “Running WebSphere Application Server with IBM Technology for
JVM” on page 167. It has detailed instructions about how to make the switch.

Chapter 4. Making the switch to IBM Technology for JVM 33

34 IBM Technology for Java Virtual Machine in IBM i5/OS
 5

Chapter 5. Tuning the garbage collector

This chapter provides details on the operation and tuning of the garbage collector component
of the IBM Technology for JVM. Garbage collection is performed differently in IBM
Technology for JVM compared to Classic JVM. Four policies are available to choose from
which provide different garbage collection characteristics. In some circumstances choosing a
different policy might provide better performance, depending on the metrics by which
application performance is measured.

IBM Technology for JVM includes an option to produce verbose output, which gives great
detail regarding the operation of the garbage collector. You can analyze this information to
give insight regarding whether adjusting the garbage collection policy and command-line
parameter settings might lead to increased garbage collection performance.

This chapter covers the following topics:

򐂰 5.1, “Introduction to garbage collection” on page 36
򐂰 5.2, “Available options” on page 37
– 5.2.1, “Garbage collection policies” on page 37
– 5.2.2, “Tuning heap size” on page 41
– 5.2.3, “Verbose GC output” on page 44
򐂰 5.3, “Choosing the right policy” on page 46

© Copyright IBM Corp. 2007. All rights reserved. 35

5.1 Introduction to garbage collection
A key differentiator of Java compared to many prominent programming languages is the
automatic allocation and deallocation of memory by the Java Virtual Machine (JVM). This
removes the burden of explicit memory management from the programmer, reducing the
scope for errors and memory leaks.

Memory management involves two distinct but related functions of the JVM:
򐂰 A memory allocator to allocate areas of storage for objects
򐂰 A garbage collector to free memory when it is no longer being used

Allocating memory is an inexpensive and speedy operation compared to reclaiming unused

space. Therefore, in discussing memory management in the JVM, we focus on the more
time-consuming process of reclaiming unused storage, commonly referred to as Garbage
Collection (GC).

When a JVM initializes, it reserves an area of system memory called the heap. As the Java
application runs and objects are created, these are stored in the heap. Because system
memory is finite, the heap also has a finite size. The heap has an initial, or minimum, size and
also a maximum size. The heap size can expand or contract automatically during the lifetime
of the JVM as required by the garbage collector.

The garbage collector must ensure that there is enough free space in the heap to satisfy new
allocation requests. It does this by marking live objects and sweeping dead objects to free
memory. These are objects to which there are no references from elsewhere in the JVM, such
as from other objects. You can then use this memory for new allocations.

If the heap runs out of space for some reason, it can cause the JVM to fail with an
OutOfMemoryException message and other messages to this effect. Therefore, garbage
collection is an important function of the JVM.

5.1.1 Garbage collection in IBM Technology for JVM

The Garbage Collection component has always been present in IBM JVMs. However the way
in which GC is implemented in the IBM Technology for JVM shipped with i5/OS V5R4, differs
compared to the Classic JVM.

In the Classic JVM, the Garbage Collector runs asynchronously (in the background) and
concurrently (at the same time) to application threads. This collector is unique to i5/OS, it
does not use “Stop The World” approach for collecting objects.

Tuning options in the Classic JVM are limited to setting the minimum heap size and maximum
heap size. Setting the minimum heap size provides rudimentary control over the frequency of
GC cycles. There is no compaction of the heap to reduce fragmentation. This means that the
heap can grow to a larger size than actually required by the application.

Fragmentation occurs because Java objects require contiguous storage. Over the lifetime of
the application, many allocation and free operations take place, each involving different size
chunks of memory. This gradually results in the heap containing many small fragments of
memory. These fragments might represent a significant chunk of memory in total, yet there is
insufficient contiguous storage to fulfill new allocation requests. Eliminating fragmentation is
therefore a key requirement of the garbage collector. GC in IBM Technology for JVM uses a
compaction technique which moves all live object to the beginning of the heap. Thus,
compaction aggregates all free chunks of memory in a single, continuous piece of memory.

36 IBM Technology for Java Virtual Machine in IBM i5/OS

 With the introduction of the IBM Technology for JVM, GC operation and tuning methods
available on i5/OS are now consistent with other platforms. The IBM Technology for JVM
defines several GC policies which enable you to tailor the way that GC interacts with your
Java application. There are four such GC policies available to choose from. These are
discussed in 5.2.1, “Garbage collection policies” on page 37

There is also scope for fine-tuning other garbage collection attributes to meet your
application’s requirements which are described in 5.2, “Available options” on page 37.

Heap size considerations

If your application currently uses the Classic JVM and you are considering switching to IBM
Technology for JVM, the heap size requirement of your application is an important
consideration. The maximum heap size with the Classic JVM is approximately 243 GB. There
is practically no application that requires such a big heap. With IBM Technology for JVM, the
maximum theoretical heap size is 4 GB, an inherent limitation of using a 32-bit address
space. In practice, expect around 2.5 GB to 3 GB actual memory to be available for the heap
due to the JVM operational requirements.

It is possible that your application running on Classic JVM uses a heap larger than 3 GB and
therefore you might think that it will not run under the IBM Technology for JVM. However,
bear in mind that object references are twice the size in a 64-bit JVM compared to a 32-bit
JVM. Therefore, object references require more storage and result in a larger memory
footprint in the Classic JVM.

Conversely, running the application under IBM Technology for JVM requires less storage for
object references and therefore the memory footprint is reduced. In this way, applications that
seemingly are not going to fit within the address space of the IBM Technology for JVM, can
due to the contraction in size of object references and other factors.

In testing, memory footprints of applications decreased by between 20% and 50% when
moved from Classic JVM to IBM Technology for JVM. Therefore, applications running under
Classic JVM with up to approximately a 4 GB heap are good candidates to run under IBM
Technology for JVM. If your current application requires up to 6 GB of memory, IBM
Technology for JVM might or might not fit this application. You have to perform stress testing
of your application running in 32-bit JVM to understand memory requirements.

Refer to 4.1.1, “Measuring heap usage in Classic JVM” on page 29 for more information
about how to determine the current heap size of the application under the Classic JVM.

5.2 Available options

This section discusses the four garbage collection policies available in IBM Technology for
JVM and also options for tuning the heap size and generating verbose garbage collector
output.

5.2.1 Garbage collection policies

Policies are used to determine the garbage collection algorithm that the JVM uses. There are
four GC policies available in IBM Technology for JVM version 5.0 and each has different
characteristics. You may ask why there is so much choice. In some circumstances the
different characteristics of a particular policy might lead to better application performance,
depending on your performance goals.

Chapter 5. Tuning the garbage collector 37

 Note: The default GC policy works well in most situations. Although changing the GC
policy is straightforward, it may have a negative impact on application performance. It is
recommended that you thoroughly test the new policy in a test environment before
considering to use the new policy in production.

The garbage collection component in IBM Technology for JVM has a variety of functions
available at its disposal, such as concurrent marking and concurrent sweeping.
Behind-the-scenes, each GC policy defines which of these functions must be used in which
combination to achieve a particular goal.

Table 5-1 lists the available policies and explains when you must use each one.

Table 5-1 GC policies available in IBM Technology for JVM

Policy Command-line parameter Description

Optimize for -Xgcpolicy:optthruput The default policy. Typically used for applications where raw
throughput (optional) throughput is more important than short GC pauses. The
application is stopped each time that garbage is collected.

Optimize for pause -Xgcpolicy:optavgpause Trades high throughput for shorter GC pauses by performing
time some garbage collection concurrently. The application is
paused for shorter periods compared to optthruput.

Generational -Xgcpolicy:gencon Handles short-lived objects differently than objects that are
concurrent long-lived. Applications that have many short-lived objects can
see shorter pause times with this policy while still producing
good throughput.

Subpooling -Xgcpolicy:subpool Uses an algorithm similar to the default policy but employs an
allocation strategy that is more suitable for multiprocessor
machines. Recommended for symmetric multiprocessing
(SMP) machines with 16 or more processors. Applications that
have to scale on large machines can benefit from this policy.

If you want to use a policy other than the default optthruput policy, you have to use the
-Xgcpolicy command line parameter with an appropriate argument when launching the JVM.
If you do not specify the -Xgcpolicy parameter, the default policy is used.

For example, to use the Optimize for pause time (optavgpause) GC policy, you use the
following command line:
java -Xgcpolicy:optavgpause HelloWorld

The following sections describe the characteristics of each policy in more detail.

Optimize for throughput (optthruput)

This is the default GC policy used in IBM Technology for JVM. It is designed to provide the
best performance for the majority of applications. It does this by focusing on application
throughput as the main indicator of performance.

The optthruput policy uses a mark-sweep-compact garbage collector. When a GC cycle

starts, all application threads are paused. Because of this the optthruput policy is known as a
“Stop The World” or STW collector.

All objects are then traced by the collector and those that are live (those objects which are still
referred to by other objects or elsewhere in the JVM) are marked.

38 IBM Technology for Java Virtual Machine in IBM i5/OS

After the mark phase is complete, the collector ‘sweeps’ objects that were not marked (which
are the dead objects), freeing the space. After the sweep phase, memory contains areas of
free space and areas occupied by live objects.

The mark and sweep phases take place during every GC cycle. At the end of the sweep
phase there may still be insufficient memory to satisfy the allocation request. This can occur if
there is excessive fragmentation of the heap.

If this scenario arises, the heap is compacted. Compaction moves objects toward the
beginning of the heap, aligning them so that no free space remains between objects as
shown in Figure 5-1. Compaction is an expensive operation compared to the mark and sweep
phases, therefore, GC avoids it unless absolutely necessary to prevent the JVM from running
out of memory.

Figure 5-1 Heap layout before and after garbage collection

Optimize for pause time (optavgpause)

The default optthruput policy can lead to varying application pause times for each GC cycle
because the mark and sweep work starts only after the application is paused.

For some applications, varying GC pause times can be undesirable. For instance, graphical
applications which require consistent response times to user interaction might not be best
suited to optthruput.

The optavgpause policy sacrifices some application throughput in order to reduce the
average GC pause time. A mark-sweep-compact collection process is still used, but the
application threads perform some of the mark and sweep work themselves while the
application is still running as shown in Figure 5-2. One or more low-priority GC threads also
run in the background to perform concurrent marking while the application is idle.

Stop-The-World collection is still used but does not take as long as optthruput because some
of the work has already been done. The throughput penalty when using optavgpause is
around 5% but this varies by application.

Figure 5-2 Distribution of CPU time in the optavgpause GC policy

Chapter 5. Tuning the garbage collector 39

 Generational concurrent (gencon)
The generational concurrent policy exploits the so-called weak generational hypothesis that
has emerged, which asserts that “most objects die young”. The gencon policy combines
concurrent mark and generational GC to minimize GC pauses. Applications that create a lot
of short-lived objects (such as transaction-based applications) might be suited to the gencon
policy.

Generational GC separates the heap into two areas referred to as nursery space and tenured
space. Allocations are initially in the nursery space. If an object survives enough garbage
collections, it is promoted to the tenured space, and the object is said to have tenured. The
nursery and tenured areas are collected differently for performance.

The nursery space is further split into an allocate space and a survivor space as shown in
Figure 5-3. Objects are allocated to the allocate space. When the allocate space is full, a GC
process called scavenge is triggered. The scavenge copies live objects into either the survivor
space, or tenured space if they are old enough. Age is measured in terms of how many
garbage collections the object has survived. The JVM determines dynamically how old an
object must be to tenure, however the maximum is 14 GC cycles. Live objects are copied to
the survivor space and aligned in such a way so that it can avoid fragmentation.

Figure 5-3 The heap is divided into nursery and tenured spaces in the gencon GC policy

After live objects have been copied into the survivor space, the allocate space contains only
dead objects. The allocate space and survivor space then swap or flip roles, this means that
the allocate space becomes the survivor space for the next scavenge and the survivor space
becomes the allocate space. On the next scavenge, live objects overwrite dead objects from
the previous scavenge for efficiency. The allocate and survivor spaces then flip roles again.
The scavenge process runs on the nursery area of the heap only, therefore, the cost of
collecting the entire heap on each GC cycle is not incurred. See Figure 5-4.

Objects in the tenured space typically live longer than those in the nursery, and therefore the
tenured space is deliberately collected less often than the nursery. The tenured space is
marked concurrently using a similar approach to the optavgpause policy, hence the policy is
called generational concurrent (gencon). However, concurrent sweep is not used, therefore,
when concurrent marking of the tenured space is completed, the application is paused for the
global collection to be completed. This global collection collects garbage in both the nursery
and tenured spaces.

Before collection
Nursery Tenured space
Allocate space Survivor space

After collection

Live objects survive the copy

Objects that have survived long enough are promoted to the tenured space

Figure 5-4 Example of heap layout before and after garbage collection using gencon policy

40 IBM Technology for Java Virtual Machine in IBM i5/OS

 Subpooling (subpool)
The subpool policy is available on IBM eServer™ System i, IBM eServer System p™, and
IBM eServer System z™. It is designed for use on multiprocessor machines (typically 16
processors or more). When scalability of your application on large machines is important,
consider using subpooling.

All GC policies in IBM Technology for JVM employ a free list. The free list is a data structure
that keeps track of which areas of the heap are available for allocations. The free list also
contains information about the size of each area of free space. Typically a single free list is
maintained for the entire heap. Much of the list might have to be searched before a suitable
chunk of free space is found.

Subpooling makes use of multiple free lists referred to as pools. A pool contains the location
of chunks of free space of a specific size. Each pool is associated with a different size as
shown in Figure 5-5. The pools are ordered by size, therefore the memory allocator only has
to go to the first element of the right pool to allocate an object. This results in faster object
allocation compared to GC policies with a single free list, because the only search required is
to locate the correct size pool to allocate from.

Subpooling also makes use of per-processor “mini heaps” which further improve performance
on multiprocessor systems. These are managed automatically by the JVM, therefore no
mechanism is provided for adjustment.

Figure 5-5 Subpooling employs multiple pools each with a list of free chunks of a specific size

5.2.2 Tuning heap size

The size of the heap available to an application can influence performance dramatically. Too
small a heap, and the garbage collector has to run frequently, reducing application throughput
and increasing response times. If the heap is much larger than required by the application
however, garbage collection takes place rarely and takes longer because the whole heap has
to be scanned for live objects. Therefore, it is important to achieve a balance. A good rule of
thumb is to have a heap 1.5 times the actual size required.

Chapter 5. Tuning the garbage collector 41

 As mentioned in “Heap size considerations” on page 37, applications use a smaller heap
under IBM Technology for JVM compared to Classic JVM. The heap minimum and maximum
size settings your application might have used with the Classic JVM will not necessarily be
valid with IBM Technology for JVM. For example, an unlimited maximum heap size is not
supported in IBM Technology for JVM. If you do not specify a maximum, the default of 2 GB is
used. For further information, refer to 4.2, “Configuring garbage collection” on page 30.

Default heap size settings in IBM Technology for JVM

With the IBM Technology for JVM you can set both the minimum and maximum size of the
heap, or you can use the default settings. On i5/OS the default minimum heap size is 4 MB
and the default maximum size is 2 GB. The minimum heap size is the smallest size the heap
is allowed to reach through heap contraction. The heap is initially the same size as the
minimum size. The maximum heap size is the largest size the heap is allowed to reach
through heap expansion. It is possible to set the minimum and maximum heap sizes to the
same value. This will have the effect that the heap neither expands nor contracts in size
during the life of the JVM.

To override the default heap size settings, you use the -Xms and -Xmx command-line
parameters for minimum and maximum respectively. Default settings are used if these
parameters are not specified.

For example, if you want to change a maximum heap size to 256 MB and use the default
value for a minimum heap size, then you use Xmx parameter:
java -Xmx256m HelloWorld

To specify a minimum heap size of 32 MB and use the default maximum heap size:
java -Xms32m HelloWorld

To specify a minimum heap size of 32 MB and maximum heap size of 256 MB:
java -Xms32m -Xmx256m HelloWorld

The IBM Technology for JVM now supports use of the G or g argument to specify the heap
size settings in gigabytes rather than megabytes as a convenience:
java -Xmx1g HelloWorld

This sets the maximum heap size to 1 GB. Only whole numbers of gigabytes are allowed,
therefore, specifying -Xmx1.5g returns an error.

If you try to specify too large a maximum heap, the java command returns the following error:
JVMJ9GC028E Option too large

Determining appropriate heap size settings

If you are switching to IBM Technology for JVM from Classic JVM, it is recommended that
you run the application with the default heap size settings initially and verbose GC output
enabled. Refer to 5.2.3, “Verbose GC output” on page 44. You can then use the following
approach to adjust the minimum and maximum heap size settings appropriately. This
approach is also valid even if the application was not previously run under Classic JVM.

The application must be run up to steady state with no load. Analyze the verbose GC output
at this stage. The current heap size, reported in the most recent entry in the verbose GC
output, gives you a rough indication of the initial size of the heap you must use (defined by the
-Xms command-line parameter).

42 IBM Technology for Java Virtual Machine in IBM i5/OS

If the value reported is much larger than the default value of 4 MB, consider using a slightly
lower than reported value for the minimum heap size setting. For example, if the reported
heap value is 192 MB, then set the minimum heap size of your JVM to 170 MB. This gives
more efficient compaction early on compared to using a large value.

Next, you must run the application under stress to determine an appropriate value for the
maximum heap size. When the application is processing the highest workload, check the
verbose GC output for the current heap size. Use a slightly larger value to set the maximum
heap size (-Xmx command-line parameter). For example, if JVM reports 650 MB as a heap
size under the highest workload, set the maximum heap size for your JVM to 680 MB. This
gives your JVM a small cushion in the production environment.

Heap expansion and contraction

Unless you fix the size of the heap at JVM startup by setting the minimum and maximum
heap sizes to the same value, the heap expands and sometimes contracts automatically as
required by the garbage collector to maintain efficient operation. The size of the heap,
however, never contracts below minimum size, and never expands beyond maximum size.
The current “top of the heap” is represented by heaptop in Figure 5-6.

Figure 5-6 Layout of the heap

For fixed-size heaps there is no concept of heaptop. For non-fixed heaps, heaptop may move
toward heapmin or heapmax, depending how much free space is in the heap after each
collection.

The garbage collector by default tries to maintain a minimum of 30% free space in the heap
for new allocations. If less than 30% free space is available after garbage has been collected
in the current GC cycle, the collector might expand the heap by the amount required to reach
30% free space. You can override the minimum free space the garbage collector tries to
maintain by specifying the -Xminf command-line parameter. Specify the size as a decimal in
the range 0-1.

For example to maintain 20% minimum free space in the heap, specify the following:
java -Xminf0.2 HelloWorld

Chapter 5. Tuning the garbage collector 43

 If there is a lot of free space in the heap after garbage has been collected, the garbage
collector may contract the heap. This can improve performance because a smaller heap can
be scanned for live objects faster. By default, if more than 60% of the heap is free space, the
garbage collector contracts the heap so that a maximum of 60% free space exists. You can
override the maximum free space by specifying the -Xmaxf command-line parameter. Specify
the size as a decimal in the range 0-1.

For example to maintain a maximum of 50% free space in the heap, specify the following:
java -Xmaxf0.5 HelloWorld

Note: As with other GC related settings, the default -Xminf and -Xmaxf settings work well,
therefore, most users do not have to adjust them.

5.2.3 Verbose GC output

IBM Technology for JVM includes an option to produce verbose GC output to give a detailed
insight into the operation of the garbage collector. It is possible to monitor the size of the heap
through analyzing verbose GC output. This enables you to determine the smallest suitable
heap size for your application and also the amount of room required for heap expansion.
Analyzing verbose GC output can show whether using a different GC policy has a positive or
negative impact on the GC characteristics of your application.

Verbose GC output produced by IBM Technology for JVM is now formatted as Extensible
Markup Language (XML). This allows tools to be written to analyze verbose GC output
automatically.

You can collect verbose GC output for review if you want to analyze the GC characteristics
and performance of an application. You can use the analysis of the information in verbose GC
output to support a decision to move to a different GC policy or to tune GC options such as
the heap size. Verbose GC output can also help you identify possible causes of poor GC
performance.

Verbose GC output is enabled by setting one of two command-line options when launching
the JVM, either -verbose:gc (or -verbosegc) or -Xverbosegclog:filename. -verbose:gc
writes its output to the standard error stream, while -Xverbosegclog writes its output to the file
filename.

For example, to enable verbose GC output and capture the output in a file GClog.txt stored in
the Integrated File System (IFS), use the following:
java -Xverbosegclog:/home/rumsbya/GClog.txt HelloWorld

Tip: To analyze the GC output with the tools, use the verbosegclog option. Tools do not
read the standard error stream correctly because it might contain some other data, for
example, exception stacks.

Interpreting verbose GC output

This section explores the following extract from a verbose GC output file shown in Figure 5-7
to identify the key elements which you can use to determine how GC is performing. For this
example we ran WebSphere Application Server V6.1.0.1 Base with an artificially small heap
size for demonstration purposes. We also started the Trade 6 performance benchmark
sample application with 500 virtual users to generate load on the server. For more information
about the Trade 6 application, refer to “Trade 6 application” on page 133.

44 IBM Technology for Java Virtual Machine in IBM i5/OS

 The results are not typical, but nevertheless they show how much useful information can be
gained from reviewing verbose GC output.

...
<af type="tenured" id="36" timestamp="Mon Oct 09 13:58:15 2006" intervalms="382.636">
<minimum requested_bytes="972016" />
<time exclusiveaccessms="0.117" />
<tenured freebytes="14877752" totalbytes="130819584" percent="11" >
<soa freebytes="14877752" totalbytes="130819584" percent="11" />
<loa freebytes="0" totalbytes="0" percent="0" />
</tenured>
<gc type="global" id="36" totalid="36" intervalms="85383.105">
<compaction movecount="885558" movebytes="53290192" reason="compact to meet allocation" />
<classloadersunloaded count="58" timetakenms="11.881" />
<refs_cleared soft="7" weak="429" phantom="152" />
<finalization objectsqueued="336" />
<timesms mark="233.177" sweep="8.977" compact="527.274" total="781.744" />
<tenured freebytes="77525984" totalbytes="130819584" percent="59" >
<soa freebytes="76217824" totalbytes="129511424" percent="58" />
<loa freebytes="1308160" totalbytes="1308160" percent="100" />
</tenured>
</gc>
<tenured freebytes="76553968" totalbytes="130819584" percent="58" >
<soa freebytes="75245808" totalbytes="129511424" percent="58" />
<loa freebytes="1308160" totalbytes="1308160" percent="100" />
</tenured>
<time totalms="782.403" />
</af>
...
Figure 5-7 Extract from verbose GC log for one collection using optthruput policy

The top-level element <af> indicates this GC cycle occurred due to an allocation failure.
Other possible values you might see are <sys> for collections forced by System.gc() calls and
<con> for concurrent collections.

Tip: <sys> elements in verbose GC output indicate the application is making calls to
System.gc(). You are strongly discouraged to use System.gc() calls to try to influence
garbage collection because this can severely degrade performance. Remove System.gc()
calls from the application or use the -Xdisableexplicitgc command-line option when
launching the JVM so that System.gc() calls have no effect.

Note the intervalms attribute in the first line:

<af type="tenured" id="36" timestamp="Mon Oct 09 13:58:15 2006"
intervalms="382.636">

This shows that only 382.636 milliseconds have passed since the last allocation failure. This
can be an indication that the heap is filling up too quickly. Consider increasing the maximum
heap size.

The next line shows the size of the allocation request that caused the failure, in this case
nearly 950 KB:
<minimum requested_bytes="972016" />

Chapter 5. Tuning the garbage collector 45

 Notice that there are three <tenured> elements for each garbage collection. These describe
the state of the heap at different stages of the GC cycle:
򐂰 The first represents the heap prior to GC, right after the allocation failure occurred.
򐂰 The second represents the heap after collection has occurred.
򐂰 The third shows heap occupancy after the request which originally caused the allocation
failure has been satisfied.

Note: It is recommended that you use the value of totalbytes from the first <tenured>
elements when you estimate the maximum heap size setting of your JVM.

The verbose GC output shows the effect of fragmentation on the heap:

<tenured freebytes="14877752" totalbytes="130819584" percent="11" >

Although there is over 14 MB total free space in the heap, there is not a large enough
contiguous chunk to allocate the requested 950 KB. Therefore, compaction is triggered to
move objects toward the beginning of the heap. On this occasion over 880,000 objects are
moved:
<compaction movecount="885558" movebytes="53290192" reason="compact to meet
allocation" />

From the verbose GC output you can see the high cost of compaction compared to the mark
and sweep phases of the GC cycle. This line shows the time in milliseconds taken for each
phase of the GC cycle:
<timesms mark="233.177" sweep="8.977" compact="527.274" total="781.744" />

On this occasion, compaction took more than twice as long than the mark phase and nearly
60 times longer than sweeping dead objects. The whole GC cycle took nearly 0.8 seconds.
This emphasizes the importance of avoiding compaction.

One approach is to accept that compaction is required at some stage in an application

lifetime, and to minimize the impact by setting a small minimum heap size. The heap requires
compaction early on, but because the heap is small it does not take as long as when the
application has reached steady state and expanded toward the maximum size. Early
compaction has the effect that most long-lived objects are at the beginning of the heap by the
time the heap starts expanding. By comparison, a large initial heap takes longer to fill and is
more fragmented by the time the first garbage collection occurs. Long-lived objects will be
scattered throughout the heap and costly to relocate to the beginning of the heap.

Another approach is to use the generational concurrent (gencon) GC policy, particularly if

your application creates a lot of short-lived objects. Because gencon uses a copying collector,
efficient compaction of the nursery occurs as a side-effect of each scavenge, versus
expensive compaction of the entire heap in other policies. However, we must point out that
the nursery collection happens very frequently.

5.3 Choosing the right policy

It is important to consider garbage collection in the context of overall application performance.
For most enterprise applications, the JVM is just one component in a system which may also
include databases, network infrastructure, Lightweight Directory Access Protocol (LDAP)
directories, Web servers, application servers, and other components. Each of these can
contribute positively or negatively to application performance. The application itself can be a
bottleneck if poorly written.

46 IBM Technology for Java Virtual Machine in IBM i5/OS

Therefore, you must consider the possible performance gains carefully before investing effort
in tuning garbage collection or adjusting the GC policy. For instance, if your Web server is
unable to keep up with the number of incoming requests, then there is little to be gained by
tuning the garbage collector. On the other hand, if you have a mature and proven application
environment that has been well-tested, you might want to consider tuning garbage collection
as part of a wider tuning strategy for your application.

Before you consider changing the garbage collection policy, you have to determine which
performance characteristics are important for your application. Throughput and response time
are two measurements of an application’s performance that you must consider. Throughput
measures how much data is processed by the application in a given time period, usually
expressed as operations or transactions per second. Response time measures how long an
application takes to complete processing of a request, starting from when the request is
received by the application.

Different applications place different importance on each of these performance

measurements. An interactive GUI-based application, for example, typically requires to have
consistently low response times. If the response time varies wildly, users are not satisfied
with the performance. A batch payroll application on the other hand, has to process a lot of
data in a given time period (high throughput), while the time taken to respond to each
individual request is of little consequence.

The default garbage collection policy is optimized to provide high application throughput while
providing GC pauses short enough for the majority of applications. It is recommended that
you run the application using the default GC policy initially. If this gives satisfactory
performance, then there is no requirement to try other GC policies, although doing so is
straightforward. Just specify the relevant -Xgcpolicy:policy command-line parameter when
launching the JVM.

Note: Before considering changing the GC policy, you must be using a heap size
appropriate to your application’s requirements. For example, if the maximum heap size is
set too small, then performance suffers regardless of which GC policy you choose.

If the default GC policy does not give satisfactory performance, analyze verbose GC output
as described in “Interpreting verbose GC output” on page 44, looking for clues to find the
particular aspect of GC behavior which is causing concern.

Tip: There is an excellent article available from IBM which presents a quantitative
approach to choosing a GC policy using a case study. The article is available from IBM
developerWorks® at:
http://www-128.ibm.com/developerworks/java/library/j-ibmjava3/

For instance, excessive fragmentation of the heap might be occurring if the application is
creating many short-lived objects. As a result, you will see in the verbose GC output that
compaction is happening too often, increasing the GC pause time. Subsequently the
application response time may increase, although the extent to which this affects the
response time depends on the workload. The effect is more noticeable in applications that are
lightly-loaded compared to heavily-loaded applications in which GC pause time makes up
only a small fraction of a response time which also includes more significant factors such as
network latency. Switching to the generational concurrent GC policy reduces fragmentation
and therefore reduces pause times.

Chapter 5. Tuning the garbage collector 47

 For interactive applications where the response time is varying unpredictably, the verbose
GC output might show that there are long pauses for GC. In this case consider using the
following policy:
optavgpause

If you have a multiprocessor server with 16 or more processors (such as a 16-way IBM
System i5™ 570 or IBM System i5 595) consider using subpooling for faster object allocation.
Per-processor “mini heaps” used automatically in this policy might further improve
performance. In addition to IBM System i5, subpooling is also available on IBM System p5™
and IBM System z9™ platforms.

48 IBM Technology for Java Virtual Machine in IBM i5/OS

 6

Chapter 6. Optimizing performance with

shared classes
This chapter discusses the new shared classes feature available in IBM Technology for JVM.
Shared classes are loaded once and shared by multiple JVMs that are running on the server.
This can improve JVM startup times if the required classes have been loaded previously by
another JVM.

This chapter presents an overview of class loading, then explains how shared classes fit into
the class loading model. The details of the shared classes featured in IBM Technology for
JVM are then explained.

The effect of shared classes on application performance is then discussed. This chapter
gives an overview of how you can integrate shared classes support into your own
custom-written classloaders and mention security considerations for using shared classes.

This chapter is divided into the following sections:

򐂰 6.1, “Introduction to shared classes” on page 50
򐂰 6.2, “Shared classes in IBM Technology for JVM” on page 53
򐂰 6.3, “Shared classes and application performance” on page 62
򐂰 6.4, “Shared Classes Helper API” on page 63
򐂰 6.5, “Security considerations” on page 67
򐂰 6.6, “Bytecode modification” on page 68

© Copyright IBM Corp. 2007. All rights reserved. 49

6.1 Introduction to shared classes
If you have never used shared classes before, you must read this section. It helps you to
comprehend information presented in the rest of this chapter.

6.1.1 Classes and classloaders

Unlike languages such as C which are linked statically at compile time, Java is a
dynamically-linked language. This means that the classes required by a Java application are
loaded into the runtime environment as required during application execution.

You must load classes into memory before you can create any instances of those classes, or
any static data referenced by other classes. You can load classes from several sources, such
as from a .class, zip, or Java archive (JAR) file present in the local filesystem, or from a
remote server.

Categories of classloader
Classes are loaded by a classloader. There are several classloaders supplied with IBM
Technology for JVM, which are used for different class-loading purposes:
򐂰 Bootstrap classloader
Loads classes from the boot class path (typically classes under jre/lib, such as classes
stored in rt.jar). These classes constitute the core of the runtime environment required by
all applications (such as Object or RuntimeException). The bootstrap class loader is
built-in to the JVM.
򐂰 Extension classloader
Loads classes from the installed extensions located on the extension class path. For
example, the IBM JCE provider classes which are included as a value-add feature of IBM
JVMs. In IBM Technology for JVM on i5/OS, such extensions are found in the following
directory by default:
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/ext
򐂰 System classloader (also known as Application classloader)
Loads classes from the general class path (java.class.path system property) such as the
classes that make up an application.

When a classloader is asked to load a class, it searches for the class in the following order:
򐂰 Classloader cache: if this classloader has previously loaded the class, it reads the class
from its cache.
򐂰 Parent classloader
򐂰 Shared class cache: if shared classes are enabled. Refer to 6.2, “Shared classes in IBM
Technology for JVM” on page 53.
򐂰 Filesystem

From this you can see a delegation model is used to load classes, as shown in Figure 6-1.
The particular delegation model used in IBM Technology for JVM is referred to as parent first
delegation. This means that if the current classloader has not already loaded the class, it asks
its parent classloader if it has already loaded the class. If the parent has not loaded the class
already, it delegates to its parent, and so on up to the bootstrap classloader.

50 IBM Technology for Java Virtual Machine in IBM i5/OS

If the class is not already loaded by any classloader, the classloader which was initially asked
to load the class, does so. The class is loaded from the filesystem, or possibly from the
shared class cache, if shared classes are enabled.

Figure 6-1 Classloader delegation model

The usage of these different classloaders is largely transparent to the programmer. The
programmer must be aware of the order in which the runtime environment searches for
classes, to ensure no name clashes occur. In practice, it is rare that any conflicts will occur,
therefore, exactly how application classes get loaded is usually of little concern unless you
are using a user-defined classloader.

User-defined classloaders
It is possible to write your own classloader in Java. This can be useful to extend class-loading
behavior and capabilities. For example your application might have to:
򐂰 Find and load a class from a unique location in the file system.
򐂰 Load a different version of the class that has been loaded by another classloader. You can
load the same class multiple times as long as each version of the class is loaded by a
different classloader.
򐂰 Generate a class on-the-fly depending on the state of your application, or modify the
bytecode to add tracing.

Custom classloaders are referred to as user-defined classloaders. User-defined classloaders

are created by extending the abstract class java.lang.ClassLoader or one of its subclasses.
This class provides implementations of various useful methods such as findClass(),
loadClass(), and defineClass(), which user-defined classloaders can override for enhanced
functionality. The system classloader is an example of a user-defined classloader (although it
is shipped with the JVM). It is implemented as java.net.URLClassLoader. WebSphere
Application Server ships with several user-defined classloaders such as a WebSphere
extensions classloader and application module classloader.

Chapter 6. Optimizing performance with shared classes 51

 Classloading and application startup
The entry point of a Java application is the public static void main (String args[])
method, which is usually defined in the main application class. As part of loading this initial
application class, there is a core set of classes that must also be loaded. The class Object, for
example, as the superclass of all other objects must be loaded before any application
classes. In fact, regardless of the JVM, there are hundreds of core classes that must be
loaded whenever any application starts.

In traditional classloader models, these core classes are loaded into the private address
space of the JVM. Therefore, if multiple JVMs are running, no JVM can see what classes
other JVMs might already have loaded. Loading multiple copies of the same core classes
clearly leads to a larger memory footprint for each JVM in this scenario. This aspect of the
Java runtime environment is the driving force behind research into technologies such as
shared classes.

6.1.2 Overview of shared classes

You can see that in the traditional class loader model, each JVM running on the system loads
the classes it requires from either the local filesystem or remotely, regardless of whether
another JVM has already loaded the same class into its private address space.

As mentioned in “Classloading and application startup” on page 52, hundreds of core classes
have to be loaded by the JVM for any application. On a server running multiple JVMs, this
leads to the same classes being loaded multiple times into each JVM. Class loading takes
time and also memory in which to store the loaded classes, therefore it is clearly inefficient to
load multiple copies of the same class into different areas of memory.

When enabled, shared classes remove this issue by maintaining a single copy of each class
loaded by the JVMs running on the server, which is shared between all JVMs that require it.
The classes are stored in shared memory which is accessible by all the JVMs running on the
server. When a classloader has to load a class which was not previously loaded by a parent
classloader, it looks in the shared cache in case another JVM has already loaded the required
class. Only if the class is not in the cache will the classloader search the filesystem.

Because there is only one copy of a class stored in memory, sharing between JVMs reduces
the memory footprint of each JVM. Retrieving a copy of the class from memory is clearly
faster than loading the class from a .class, zip, or JAR file stored on disk or remotely.

Another benefit of using shared classes is that the class cache remains on the system until
next initial program load (IPL). Therefore, even if you run only one JVM at a time but use
frequent restarts of a JVM, you still benefit from using the shared classes feature.

6.1.3 History of shared classes

Shared classes as a concept has been around for some time, with forms of class sharing
appearing in some IBM JVMs since Java 1.3. The IBM Persistent Reusable JVM on IBM
z/OS® was one of the first available JVMs from IBM which included a form of class sharing,
although it required JVMs to be configured in a master/slave scenario. The Classic JVM on
i5/OS employs a Java cache (implemented as a shared JAR file) available to user-defined
classloaders, allowing reuse of JVAPGM objects and class bytecodes.

Version 5.0 of IBM Technology for JVM offers, for the first time across the major platforms, a
completely transparent and dynamic means of sharing all loaded classes (with the exception
of Sun Solaris™ and the HP hybrids). Furthermore, no restrictions are placed on JVMs that
are sharing the class data.

52 IBM Technology for Java Virtual Machine in IBM i5/OS

 As explained later, however, in 6.6, “Bytecode modification” on page 68, care must be
exercised if you are using runtime bytecode modification of classes.

6.1.4 Shared classes and user-defined classloaders

In version 5.0 of IBM Technology for JVM, classes loaded by user-defined classloaders can
be shared. The class java.net.URLClassLoader has been enhanced to support shared
classes. Therefore, if a user-defined classloader extends this class, it will have shared
classes support built-in by default due to inheritance.

IBM has provided a Shared Classes Helper application programming interface (API) so that
support for shared classes can be added to user-defined classloaders which do not inherit
from java.net.URLClassLoader. The helper API is described in more detail in 6.4, “Shared
Classes Helper API” on page 63.

6.2 Shared classes in IBM Technology for JVM

This section discusses in more detail the implementation and usage of the shared classes
feature that you can enable in IBM Technology for JVM.

6.2.1 The class cache

In IBM Technology for JVM, shared classes are facilitated by one or more caches stored in
shared memory. There can be multiple caches on the server, but each JVM can connect to
only a single cache in its lifetime. When shared classes are enabled, classloaders store
classes in a cache for use by other JVMs and find classes in the cache. The cache supports
multiple JVMs reading/writing concurrently.

In addition to a cache, the first JVM to start with shared classes enabled creates
the/tmp/javasharedresources directory. This directory holds information about all the caches
on the server. Because the cache is stored in shared memory, a cache can persist only until
the next IPL, at which time it is lost. One or all caches can be explicitly destroyed at any time
if required, without the necessity to IPL.

A cache is created with a fixed size which cannot change after the cache is created. After the
cache is full, classes can still be loaded from the cache, but no more classes can be stored.
The default cache size on i5/OS is 16 MB. This can be overridden at runtime using the
-Xscmx command-line parameter. Refer to “Setting class cache size” on page 59.

The cache only contains immutable (read-only) data from classes, such as fields declared as
final static. Such data is stored in the cache as an internal structure called a ROMClass. Data
which could change is stored separately in a structure called a RAMClass. They are stored in
the JVMs private process memory and point to the relevant ROMClass. Multiple RAMClasses
can point to the same ROMClass because the ROMClass contains read-only data. This is the
fundamental principle on which class sharing is based. The distinction between ROMClass
and RAMClass is managed by the JVM and is transparent to classloaders.

Metadata is stored in the cache along with the ROMClass consisting of versioning information
and information about where each class was loaded from. This is used to make sure the
correct class is loaded from the cache in cases where the on-disk representation of the class
changes during the cache lifetime or if the same class exists on a different class path.

Chapter 6. Optimizing performance with shared classes 53

 When shared classes are enabled, the delegation model outlined in “Categories of
classloader” on page 50 still applies. This means that parent classloaders up to the bootstrap
classloader are asked in turn for the class in case they have already loaded it. If no parent
classloader has loaded the class previously, rather than going straight to the filesystem to
locate the class, the shared classes cache is searched. Only if the class is not found in the
cache will the filesystem be searched. When the class is found, it is added to the class cache,
if the cache is not full.

Shared classes scope

All classes loaded by the runtime environment, both system and application, can be shared.
The bootstrap, extension, and system classloaders have been modified to take advantage of
the shared classes feature when it is enabled. Custom classloaders have to either extend
java.net.URLClassLoader or use the Shared Classes Helper API to integrate shared classes
features.

WebSphere Application Server and shared classes

Version 6.1 of WebSphere Application Server uses the shared classes feature automatically
and transparently to improve performance.

6.2.2 Deploying shared classes

The shared classes feature is disabled by default in IBM Technology for JVM. Shared classes
are enabled and managed through use of the -Xshareclasses command-line parameter and
its suboptions.

Important: It is recommended that you apply program temporary fix (PTF) SI25920. This
fixes an issue whereby a class name had to be specified for -Xshareclasses suboptions to
work, even for the utility options.

Enabling shared classes

Shared classes are enabled implicitly by creating at least one class cache for JVMs to store
shared classes in. To create a new class cache, the name= suboption is used when starting an
application. If a cache of the same name already exists, the JVM connects to the existing
cache. Figure 6-2 is an example which creates two caches.

QSH Command Entry

> java -Xshareclasses:name=cache1 HelloWorld

Hello world!
$
> java -Xshareclasses:name=cache2 HelloWorld
Hello world!
$

===> _________________________________________________________________________________________
______________________________________________________________________________________________
______________________________________________________________________________________________
______________________________________________________________________________________________
F3=Exit F6=Print F9=Retrieve F12=Disconnect
F13=Clear F17=Top F18=Bottom F21=CL command entry
Figure 6-2 Creating two class caches

54 IBM Technology for Java Virtual Machine in IBM i5/OS

 Listing all class caches
To show the caches that exist on the server, use the listAllCaches suboption as shown in
Figure 6-3.

QSH Command Entry

> java -Xshareclasses:listAllCaches

Shared Cache OS shmid in use Last detach time
cache2 5833 0 Wed Sep 27 16:05:23 2006

cache1 5831 0 Wed Sep 27 16:05:12 2006

Unable to create Java Virtual Machine.

$

===> ________________________________________________________________________________________
______________________________________________________________________________________________
______________________________________________________________________________________________
______________________________________________________________________________________________
F3=Exit F6=Print F9=Retrieve F12=Disconnect
F13=Clear F17=Top F18=Bottom F21=CL command entry
Figure 6-3 The listAllCaches suboption lists all the class caches present on the system

Note: Some of the -Xshareclasses suboptions are utility options. Although they use the
JVM launcher (the java command), by design a JVM is not started. Consequently, an
“Unable to create Java Virtual Machine” message is always output by the launcher
after a utility has run, as shown in Figure 6-3. This message is not an error.

Chapter 6. Optimizing performance with shared classes 55

 Viewing summary class cache statistics
To view summary statistics for a named cache, use the printStats suboption as shown in
Figure 6-4.

QSH Command Entry

> java -Xshareclasses:name=cache1,printStats

Current statistics for cache "cache1":

base address = 0xE0000058

end address = 0xE0FFFFF8
allocation pointer = 0xE010F4A0

cache size = 16777128

free bytes = 15652864
ROMClass bytes = 1111112
Metadata bytes = 13152
Metadata % used = 1%

# ROMClasses = 278
# Classpaths = 2
# URLs = 0
# Tokens = 0
# Stale classes = 0
% Stale classes = 0%

Cache is 6% full

Unable to create Java Virtual Machine.

$

===> ________________________________________________________________________________________
______________________________________________________________________________________________
______________________________________________________________________________________________
______________________________________________________________________________________________
F3=Exit F6=Print F9=Retrieve F12=Disconnect
F13=Clear F17=Top F18=Bottom F21=CL command entry
Figure 6-4 Displaying summary cache statistics using printStats suboption

56 IBM Technology for Java Virtual Machine in IBM i5/OS

 Viewing class cache contents
The printAllStats suboption lists the entire contents of the cache, and also displays the
summary statistical information returned by the printStats suboption. The output is listed in
the chronological order by which classes are added to the cache, therefore you can use them
as an “audit trail”. Each JVM that connects to the cache receives a unique ID which is shown
preceding each entry in Figure 6-5.

QSH Command Entry

> java -Xshareclasses:name=cache1,printAllStats

Current statistics for cache "cache1":

1: 0xE0FFF8B4 CLASSPATH
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/vm.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/core.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/charsets.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/graphics.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/security.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/ibmpkcs.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/ibmorb.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/ibmcfw.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/ibmorbapi.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/ibmjcefw.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/ibmjgssprovider.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/ibmjsseprovider2.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/ibmjaaslm.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/ibmcertpathprovider.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/server.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/xml.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/IBMmisc.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/lib/IBMi5OSJSSE.jar
1: 0xE0FFF87C ROMCLASS: java/lang/Object at 0xE0000058.
Index 0 in classpath 0xE0FFF8B4
1: 0xE0FFF854 ROMCLASS: java/lang/J9VMInternals at 0xE00006E0.
Index 0 in classpath 0xE0FFF8B4
1: 0xE0FFF82C ROMCLASS: java/lang/Class at 0xE00020A8.
Index 0 in classpath 0xE0FFF8B4
...
Figure 6-5 Displaying class cache contents using PrintAllStats suboption

Interpreting printAllStats output

The output shown in Figure 6-5 indicates that JVM 1 (the first JVM to use this cache) caused
a classpath to be stored at address 0xE0FFF8B4 in the cache. The classpath contains 18
entries, which are listed. If the classpath was stored using a given modification context or
partition (refer to 6.6, “Bytecode modification” on page 68) this information is also displayed.

Taking a closer look at the meaning of the address information given for each ROMClass in
the cache, in this case java.lang.Object:
1: 0xE0FFF87C ROMCLASS: java/lang/Object at 0xE0000058.
Index 0 in classpath 0xE0FFF8B4

Chapter 6. Optimizing performance with shared classes 57

 The first address, 0xE0FFF87C, is the address where the metadata for this ROMClass is
stored. The second address 0xE0000058 is the address where the ROMClass itself is located
in the cache. The next line gives the location of the classpath from which this class was
loaded (0xE0FFF8B4) and the index of this class in the classpath, which is 0 in this case. The
classpath is zero-indexed, therefore, 0 corresponds to the first entry in the classpath in
Figure 6-5. This means that java.lang.Object was loaded from the following as expected:
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/vm.jar

Deleting the class cache

It is possible to delete one or all caches at any time using the destroy and destroyAll utility
suboptions respectively, as shown in Figure 6-6.

QSH Command Entry

> java -Xshareclasses:name=cache1,destroy

JVMSHRC010I Shared Cache "cache1" is destroyed
Unable to create Java Virtual Machine.
$
> java -Xshareclasses:listAllCaches
Shared Cache OS shmid in use Last detach time
cache2 5833 0 Wed Sep 27 16:05:23 2006

Unable to create Java Virtual Machine.

$

===> ________________________________________________________________________________________
______________________________________________________________________________________________
______________________________________________________________________________________________
______________________________________________________________________________________________
F3=Exit F6=Print F9=Retrieve F12=Disconnect
F13=Clear F17=Top F18=Bottom F21=CL command entry
Figure 6-6 Deleting a class cache

There might be several reasons why you want to delete a class cache. One example is if you
want to recreate a class cache with a different size. Another example is if you do a lot of
testing with the different versions of your applications, your class cache may contain many
classes that you do not require anymore. In order to clean the cache, you delete it. Next time
you start your applications, the JVM creates a clean class cache.

Important: Only the user profile under which a cache was created, may delete the cache.
This is true even if the groupAccess suboption was specified when the cache was created.
Refer to 6.5.1, “Operating system security” on page 67 for more information about the
groupAccess suboption.

58 IBM Technology for Java Virtual Machine in IBM i5/OS

 Setting class cache size
The cache size can be set using the -Xscmx command-line parameter, as shown in
Figure 6-7.

Notice the use of the verbose suboption to show that the class cache has become full when
using an artificially small cache size of 10 KB.

QSH Command Entry

> java -Xscmx10k -Xshareclasses:name=cache1,verbose HelloWorld

^-Xshareclasses verbose output enabled]
JVMSHRC159I Successfully opened shared class cache "cache1"
JVMSHRC166I Attached to cache "cache1", size=10200 bytes
JVMSHRC096I WARNING: Shared Cache "cache1" is full. Use -Xscmx to set cache size.
Hello world!
JVMSHRC168I Total shared class bytes read=1672. Total bytes stored=0
$

===> ________________________________________________________________________________________
______________________________________________________________________________________________
______________________________________________________________________________________________
______________________________________________________________________________________________
F3=Exit F6=Print F9=Retrieve F12=Disconnect
F13=Clear F17=Top F18=Bottom F21=CL command entry
Figure 6-7 Setting class cache size using -Xscmx parameter

If you try to use -Xscmx to change the size of a cache after it has been created, the new size
is ignored.

Dynamic cache update

Suppose that the HelloWorld class is modified to include another method (which simply prints
‘doing stuff’) and recompiled.

The existing .class file in the filesystem is replaced with the updated class, invalidating the
version stored in the cache.

If you run the application again, you will see that the cache detects a newer version of the
class is available, and marks the cached copy as stale.

Chapter 6. Optimizing performance with shared classes 59

 The new class is read from disk and a copy stored in the cache, as shown in Figure 6-8.

QSH Command Entry

> java -Xshareclasses:name=myCache HelloWorld

Hello world!
doing stuff
$
> java -Xshareclasses:name=myCache,printAllStats
...
3: 0xE0FF99BC CLASSPATH
/home/rumsbya
3: 0xE0FF9984 ROMCLASS: HelloWorld at 0xE010DC10.!STALE!
Index 0 in classpath 0xE0FF99BC
...
9: 0xE0FEFA84 ROMCLASS: HelloWorld at 0xE02306D0.
Index 0 in classpath 0xE0FF99BC
...
# Stale classes = 1
% Stale classes = 0%

Cache is 14% full

Unable to create Java Virtual Machine.

$
Figure 6-8 Classes are marked as stale when a newer version is detected

6.2.3 Recommendations for shared classes

This section presents some recommendations you must consider before deploying shared
classes. It is not wise to enable shared classes without first considering the applications that
run on the system and how they are going to have to interact with the class cache.

Cache naming
If multiple users are running the same application, you may want them to use different
caches. In this case, each user can use the default cache name, which incorporates the
current user profile name to ensure uniqueness.

Or if you want a more meaningful unique name, you can use the %u modifier when specifying
the cache name, which substitutes the user profile name.

If the same operating system group of users use the same application, you might want them
to share the same cache to maximize the number of classes that are shared.

On i5/OS, by default, caches are only accessible by the user profile that created them. To
share a cache between a group of users, they must belong to the same primary group profile.

60 IBM Technology for Java Virtual Machine in IBM i5/OS

The cache must also be created with the groupAccess suboption, as shown in Figure 6-9.

QSH Command Entry

> java -Xshareclasses:name=groupTest,groupAccess HelloWorld

Hello world!
$

===> __________________________________________________________________________
_______________________________________________________________________________
_______________________________________________________________________________
_______________________________________________________________________________
F3=Exit F6=Print F9=Retrieve F12=Disconnect
F13=Clear F17=Top F18=Bottom F21=CL command entry
Figure 6-9 The groupAccess suboption allows different users in the same group to share a cache

If multiple operating system groups are running the same application, the %g modifier can be
added to the cache name. Each group running the application then gets a separate cache.

Number of caches to use

It is often the case that different applications (those in which not many classes are common)
benefit more from sharing a single larger cache than using multiple smaller caches. With a
single larger cache, you are still sharing the bootstrap classes between all JVMs, which is a
worthwhile saving in the majority of cases.

It is recommended, therefore, to keep the number of caches to a minimum to maximize the

opportunities for sharing. The exception is when separate caches are required for
maintenance purposes or to provide a degree of isolation. This may be desirable, for
example, when testing a new version of an application.

Even in these situations, it might be more efficient to create a unique modification context for
the application in the single cache, rather than to create a separate cache. Refer to
“Modification contexts” on page 68 for information about creating a modification context.

Use JAR or zip files, not .class files

The classes in the cache are automatically kept up-to-date by constant checking to ensure
the time stamp information for the class matches the filesystem. If the version of the class in
the filesystem is found to be newer, the copy in the cache is marked as stale to prevent other
JVMs loading the old class from the cache. This checking is an unavoidable source of
overhead which varies depending on whether the classloader loads most classes from JAR
or zip files, or if the majority of classes are read from .class files.

JAR and zip files are handled differently by the cache compared to .class files. When a
classloader loads a class from a JAR or zip file, it can lock the file, preventing it from being
updated. Therefore, checking is not performed as often for .class files, which because they
are not locked by the classloader, might disappear from the filesystem at any time (or a newer
version of the .class file could open, invalidating the cached version). The implication is that
cache operation is more efficient when classes are loaded from a JAR or zip file, compared to
a .class file. Therefore, when designing Java applications to exploit shared classes, consider
packaging classes into JAR or zip files rather than a number of stand-alone .class files.

Chapter 6. Optimizing performance with shared classes 61

 Cache size
The shared classes cache is a fixed size, although this size is configurable when the cache is
created. A full cache does not affect the functionality of any JVMs that are connected at the
time, or that subsequently connect to it. However if the cache becomes full it might not give
optimal efficiency or performance. Classes that are already stored in the cache continue to be
loaded from the cache. However, no new classes can be stored in the cache and therefore
they are read from disk into the JVM’s private process memory.

It is also worth knowing that if a class in a full cache is marked as stale, there is going to be no
space to store updated metadata about the class. Classes are pessimistically marked stale
and a new piece of metadata is stored if it eventually proves not to be stale, redeeming the
class. If there is no space in the cache to store the updated metadata, the cache cannot
redeem any stale classes, and they are read from disk even though the class has not
changed from the version originally stored in the cache.

We suggest, therefore, to create the cache with sufficient size for the applications that use it
so that it does not fill completely. You can do this by running the application initially using a
large cache size (set using the -Xscmx command-line parameter) then using the printStats
utility suboption to determine how much class data was stored. You must perform this after
the application has terminated because classes are loaded throughout the JVM lifetime, even
during the shutdown process. You must add a small amount to the value shown in printStats
output for contingency, as shown in Figure 6-4 on page 56. A good starting number is 10% of
the cache size, but you still have to test your application to make sure you have sufficient
cache size.

6.3 Shared classes and application performance

In addition to understanding the mechanism of the shared classes, you must consider the
type of application you are running with shared classes.

6.3.1 When to use shared classes

Shared classes leads to the greatest performance benefit when multiple JVMs run
concurrently on a system, or if JVMs restart frequently. For example, WebSphere Application
Server V6.1 uses shared classes by default. If you run multiple instances of WebSphere
Application Server on one server in a vertical cluster, you might notice the faster startup times
and less virtual memory required by each JVM.

If you are in a test or development environment and are restarting the JVM often to test
different tuning strategies, shared classes lead to faster JVM restarts.

6.3.2 Overheads associated with shared classes

The first JVM to run has to populate the cache so that classes can be shared. This adds a
slight overhead to the startup time of the first JVM, varying by the number of classes present
in the application. Expect up to 5% performance impact compared to a non-sharing JVM.
However, this initial cost can be significantly smaller compared to the time saved by
subsequent JVMs that load classes from the cache instead of from disk.

Class cache takes some space available in a JVM process. This is one of the reasons why it
is said that the practical memory limit for a Java heap is 2.5 GB to 3 GB. For WebSphere
Application Server, the maximum memory available for a Java heap is approximately 2.5 GB.

62 IBM Technology for Java Virtual Machine in IBM i5/OS

 Shared classes and garbage collection
Shared classes do not affect the operation of the garbage collector, or the eligibility of
classloaders to be garbage collected. Classloaders which use shared classes may still be
garbage collected if there are no references to the classloader instance from anywhere else
in the JVM. Class data stored as a ROMClass in the cache is never garbage collected, even if
the classloader that loaded it into the cache is collected.

6.3.3 Performance gains with shared classes

We estimate the JVM startup time for a system using shared classes improves 10% to 40%
depending on the application. The more JVMs running concurrently, the greater the
performance benefits. If the cache is being shared by similar applications, there is more
scope for sharing. Improved startup time and reduced memory footprint make a compelling
case for enabling shared classes in systems using multiple JVMs or where JVMs are
frequently restarted.

6.4 Shared Classes Helper API

IBM has provided a Shared Classes Helper API to allow developers to integrate class-sharing
support into their own classloaders. Although the details of managing shared classes are
largely handled by the JVM for you, if you are using the Helper API you have to manage
finding and storing classes in the cache using the Helper API.

Note: This section applies to you only if you plan to write your own classloader.

6.4.1 Helper API overview

The Helper API classes are in the com.ibm.oti.shared package, contained in the vm.jar file
under the following directory:
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib

Tip: It is recommended extending java.net.URLClassLoader to take advantage of shared

classes support built-in to this class. Otherwise you have to use the helper API to add
shared classes support to your custom classloader.

For your classloader to share classes, it must obtain a SharedClassHelper object from a
SharedClassHelperFactory. The SharedClassHelperFactory is a singleton object returned by
calling the following static method:
com.ibm.oti.shared.Shared.getSharedClassHelperFactory()

This method returns a factory if shared classes is enabled in the JVM. Otherwise it returns
null.

The SharedClassHelper gives the classloader a simple API for finding and storing classes in
the class cache to which the JVM is connected. After it is created, the SharedClassHelper
belongs to the classloader that requested it and can only store classes defined by that
classloader. The classloader and SharedClassHelper have a one-to-one relationship.

Calling the static findHelperForClassLoader() method of the factory returns the

SharedClassHelper object for the specified classloader, if one already exists. If the
classloader is garbage collected, its SharedClassHelper object is also garbage collected.

Chapter 6. Optimizing performance with shared classes 63

 Types of SharedClassHelper
There are several types of SharedClassHelper that you can use depending on the classpath
format the classloader has to work with. These are defined as subinterfaces of
SharedClassHelper:
򐂰 SharedClassTokenHelper
Used by classloaders that require complete control over cache contents. The classloader
stores and finds classes using an arbitrary String token generated by the classloader. This
provides greatest flexibility. However, the classloader requires additional logic to generate
and keep track of the tokens so that it can retrieve classes from the cache at a later time.
򐂰 SharedClassURLHelper
Used by classloaders that do not have the concept of a classpath. The classloader stores
and finds classes using a URL path such as a file system location. For example as follows:
file:///home/rumsbya/javaclasses
򐂰 SharedClassURLClasspathHelper
Used by classloaders that have the concept of a classpath. The classloader stores and
finds classes using a URL classpath. Each classpath is composed of one or more URLs.

Restriction: You can find classes stored by a SharedClassURLHelper using a

SharedClassURLClasspathHelper, and the opposite also applies. However, you can
find classes stored using a SharedClassTokenHelper only by using a
SharedClassTokenHelper.

Classes stored in the cache by a SharedClassTokenHelper are not dynamically updated if the
class changes in the filesystem. The reason is that the Tokens used to store classes have no
meaning to the cache, therefore, it has no way of obtaining version information to use for
comparison. Therefore, if you require cached classes dynamically updated to reflect any
changes, use a SharedClassURLHelper or SharedClassURLClasspathHelper to store
classes instead.

6.4.2 Helper API usage

Regardless of the SharedClassHelper API your classloader is using, two key functions are
provided to the classloader for working with the cache:
򐂰 findSharedClass()
The delegation model must continue to be followed by user-defined classloaders. The
classloader must continue to ask its parent for a class before checking the cache, by
calling super.loadClass().
If this returns null, indicating no parent classloader has loaded the class already, then the
classloader must call findSharedClass() on the SharedClassHelper. This returns the
ROMClass from the cache as a byte array, or null if the class is not in the cache.
Due to the distinction between ROMClasses and .class files, the byte array returned by
findSharedClass() is not the actual class bytes. The classloader cannot instrument or
modify these raw bytes in the same way that is possible with class bytes loaded from disk,
because after they are stored in the cache, they are read-only. The byte array returned by
this method must be passed to defineClass() in order to define the class for the current
JVM and return it as a java.lang.Class object.
If findSharedClass() returns null, the class is not in the cache, or the cache contains a
stale copy, which means that the cache is aware a newer version of the class is available.

64 IBM Technology for Java Virtual Machine in IBM i5/OS

 In either case, the classloader must retrieve the class bytes from disk, define the class
using defineClass(), then pass the resulting java.lang.Class object to storeSharedClass().
򐂰 storeSharedClass()
This Helper API method can be used by the classloader to store java.lang.Class objects in
the cache so that they can be shared between JVMs. The classloader must define the
class first by calling defineClass().

Important: Since the bytes returned from findSharedClass() are not actual class bytes,
the classloader must never try to define a class from these bytes and store the resulting
class in the cache using storeSharedClass(). Only classes loaded from disk must be
passed to storeSharedClass().

6.4.3 Shared Classes Helper API example

Example 6-1 shows a partial listing for a simple user-defined classloader that makes use of
the Shared Classes Helper API to work with the cache.

Example 6-1 Partial source listing for YourcoClassLoader.java

import sun.misc.URLClassPath;
import java.net.URL;
import java.lang.ClassLoader;
import java.io.IOException;
import com.ibm.oti.shared.*; // shared classes support import statement

public class YourcoClassLoader extends ClassLoader {

private URLClassPath ucp; // The search path for classes
private SharedClassURLClasspathHelper helper; // Shared classes helper object
int storeAt = 0; // current index into the cache used for storing new classes

// Wrapper class used for finding shared classes

private class SharedClassIndexHolder implements SharedClassURLClasspathHelper.IndexHolder {
int index;

public void setIndex(int index) {

this.index = index;
}
}

protected void addURL(URL url) { // manage the classpath

ucp.addURL(url);
}

private void initSharedClassesSupport(URL[] initialClassPath) {

SharedClassHelperFactory factory = Shared.getSharedClassHelperFactory();

if (factory != null) {
try {
this.helper = factory.getURLClasspathHelper(this, initialClassPath);
} catch (HelperAlreadyDefinedException ex) {
ex.printStackTrace();
}
}
}

Chapter 6. Optimizing performance with shared classes 65

 public YourcoClassLoader(URL[] urls, ClassLoader parent) {
super(parent);
if(Shared.isSharingEnabled()){
initSharedClassesSupport(urls);
}
ucp = new URLClassPath(urls, null, helper);
}

protected Class<?> findClass(final String name){

// Try to find the class from the shared cache using the class name
Class clazz = null;

if (Shared.isSharingEnabled()) {
SharedClassIndexHolder indexHolder = new SharedClassIndexHolder();
byte[] sharedClazz = helper.findSharedClass(name, indexHolder);

if (sharedClazz != null) {
clazz = super.defineClass(name,sharedClazz,0,sharedClazz.length);
return clazz;
}
else{
byte[] newClazzBytes = ... // read class bytes from disk
Class newClazz = super.defineClass(name, newClazzBytes,0,newClazzBytes.length);
helper.storeSharedClass(newClazz, storeAt);
storeAt++;
}
}
else{ // shared classes disabled, load class from disk, no cache
...
}
return clazz;
}

public final synchronized Class loadClass(String name, boolean resolve) throws

ClassNotFoundException{
return super.loadClass(name, resolve);
}
}

6.4.4 Shared Classes Helper API Javadoc

The Shared Classes Helper API documentation is shipped as a zip file with IBM Technology
for JVM on most platforms. i5/OS users can obtain this zip file by visiting the FTP site for this
IBM Redbook. Refer to Appendix E, “Additional material” on page 199 for more information
about accessing the FTP site.

The Shared Classes Helper API documentation is available in the SharedClasses folder of
the extracted apidoc.zip file. Refer to Appendix E, “Additional material” on page 199 to
download this file.

66 IBM Technology for Java Virtual Machine in IBM i5/OS

6.5 Security considerations
Using shared classes is no less secure than each JVM having its own copy of a class. Shared
classes are subject to the same bytecode verification as non-shared classes. Care has to be
taken if you are using bytecode instrumentation of classes. Refer to 6.6, “Bytecode
modification” on page 68.

Using the shared classes feature is subject to two levels of security:

򐂰 Operating system security
򐂰 Java security (only if you are using a Java SecurityManager)

6.5.1 Operating system security

Operating system security can affect shared classes in several ways. When enabling shared
classes, the first JVM to start asks the operating system to create several objects:
򐂰 A shared memory area for the cache
򐂰 A shared semaphore to coordinate access to the cache
򐂰 The following directory and two files underneath this directory for each cache:
/tmp/javasharedresources

All of these operations are subject to standard operating system security. Therefore any could
conceivably fail for a variety of reasons, such as insufficient authority or even insufficient
shared memory/disk space.

The cache is created with user access by default unless the groupAccess suboption of
-Xshareclasses is used, as shown in Figure 6-9 on page 61. This means that by default, only
JVMs started under the same user profile as the JVM which initially created the cache, is able
to use the cache.

If JVMs are started under different user profiles, these profiles have to share the same
primary group profile in order to share the same cache. Usually however JVMs start under the
same user profile, therefore, it is rare that you have to use the groupAccess suboption to
share the same cache. For example, WebSphere Application Server, by default, runs all
application servers under the QEJBSVR user profile.

6.5.2 Java security

If a Java SecurityManager is in use and the application uses custom classloaders, then these
must be granted shared class permissions before they can share classes. You modify the
java.policy file to add the shared class permissions using the class name of the custom
classloader, or through specifying a wildcard. The type of access granted to the shared
classes cache is one of “read”, “write”, or “read/write”.

Example 6-2 shows granting “read/write” shared classes permissions to all classloaders in
the com.yourco.customclassloaders package.

Example 6-2 Extract from java.policy file showing shared class permission
permission com.ibm.oti.shared.SharedClassPermission
"com.yourco.customclassloaders.*", "read,write";

Restriction: You cannot use a SecurityManager to restrict or modify the shared classes
permissions of the Bootstrap, Extension, or System classloaders supplied with IBM
Technology for JVM.

Chapter 6. Optimizing performance with shared classes 67

6.6 Bytecode modification
Runtime modification of Java class bytecode is being used increasingly often. You can use
such instrumentation to introduce additional functionality into Java classes dynamically, in
response to application state. For example, you can add sophisticated trace functionality to a
class as required (and subsequently remove it when finished with it) while the application is
still running. This reduces the overhead of having permanent trace code in all classes that
has to be controlled by a global trace variable.

You can perform bytecode instrumentation by a variety of methods. A JVM Tools Interface
(JVMTI) agent might be used to hook into the application, or the classloader itself may modify
the class bytes before defining the class. Storing such classes in the cache without any
indication they have been modified could cause major problems for JVMs that expect to
retrieve the original class from the cache.

Fortunately, IBM Technology for JVM manages the cache so that it may be shared safely by
both JVMs that use the modified class, and also those that do not.

Modification contexts
When starting the JVM, if you want to intentionally share or use a shared instrumented class,
you can specify the appropriate modification context. A modification context is a user-defined
descriptor which indicates the type of modification applied to the class bytecodes. Each
modification context is associated with a private storage area where modified classes are
stored. Specifying a modification context means that JVMs not using the modified class can
safely use the same cache as those that are.

A cache can contain multiple modification contexts. You might have a modification context
used for debugging, which is specified when starting an application that is causing problems.
If you decide to enable debugging dynamically at runtime, any classes instrumented with
debug code are safely stored in a separate area of the cache so that they cannot interfere
with the unmodified version used by other JVMs. The application being debugged still gets
the benefit of sharing any unmodified classes while it is running in debug mode.

Bytecode instrumentation might have been performed by a JVM Tool Interface agent, or by a
user-defined classloader without using a JVMTI agent. If instrumentation is performed by a
JVMTI agent, a modification context does not strictly have to be specified when sharing
classes. The JVM can detect the modification and handle partitioning transparently.

Tip: It is recommended always specifying a modification context when sharing classes that
have been modified by a JVMTI agent. Although not strictly required for safe sharing of
modified and unmodified classes in the same cache, the JVM has to perform extra checks
at runtime on modified classes not associated with a modification context. These checks
impact performance slightly.

Important: If class modification was not performed by a JVMTI agent (for example, by a
user-defined classloader) you must specify a modification context. If you do not, the cache
is not partitioned and all JVMs connected to the same cache are going to use the modified
version of the class, causing unexpected and undesirable results.

68 IBM Technology for Java Virtual Machine in IBM i5/OS

Modification context example
Modification contexts are specified by the user at runtime using the modified=context
suboption of -Xshareclasses. If a modification context of the same name already exists in the
specified cache, the JVM shares this context with other JVMs started under the same
modification context. This allows you to intentionally share modified classes between multiple
JVMs, for example, instrumented debug versions of the classes could persist across multiple
debug runs of an application for efficiency.

We used the printAllStats suboption to view the modification contexts associated with a
cache. Refer to “Interpreting printAllStats output” on page 57 for more information.

Figure 6-10 shows the HelloWorld application creating a cache called myCache and a
modification context called myModification1.

QSH Command Entry

> java -Xshareclasses:name=myCache,modified=myModification1 HelloWorld

Hello world!
$
> java -Xshareclasses:name=myCache,printAllStats

Current statistics for cache "myCache":

1: 0xE0FFF8A4 CLASSPATH
(modContext=myModification1)
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/vm.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/core.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/charsets.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/graphics.jar
...
1: 0xE0FFCD08 CLASSPATH
(modContext=myModification1)
/home/rumsbya
1: 0xE0FFCCD0 ROMCLASS: HelloWorld at 0xE010DC10.
Index 0 in classpath 0xE0FFCD08
...
Figure 6-10 Creating a new cache and modification context

The output in Figure 6-10 indicates that JVM 1 stored the class HelloWorld in the cache. The
metadata about the class is stored at address 0xE0FFCCD0 and the class itself is written to
address 0xE010DC10 in the cache. It also indicates the classpath against which the class is
stored and from which index in that classpath the class was loaded, in this case from the first
(and only) entry in the classpath at 0xE0FFCD08.

Chapter 6. Optimizing performance with shared classes 69

 If you run the application several more times, without modifying the class bytes, you can see
that only one ROMClass is stored in the cache (at 0xE010DC10) and shared between all JVMs,
as shown in Figure 6-11.

QSH Command Entry

> java -Xshareclasses:name=myCache HelloWorld

Hello world!
$
> java -Xshareclasses:name=myCache,modified=myModification2 HelloWorld
Hello world!
$
> java -Xshareclasses:name=myCache,printAllStats

Current statistics for cache "myCache":

1: 0xE0FFF8A4 CLASSPATH
(modContext=myModification1)
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/vm.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/core.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/charsets.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/graphics.jar
...
1: 0xE0FFCD08 CLASSPATH
(modContext=myModification1)
/home/rumsbya
1: 0xE0FFCCD0 ROMCLASS: HelloWorld at 0xE010DC10.
Index 0 in classpath 0xE0FFCD08
...
3: 0xE0FFC534 CLASSPATH
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/vm.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/core.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/charsets.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/graphics.jar
...
3: 0xE0FF99BC CLASSPATH
/home/rumsbya
3: 0xE0FF9984 ROMCLASS: HelloWorld at 0xE010DC10.
Index 0 in classpath 0xE0FF99BC
...
5: 0xE0FF91D8 CLASSPATH
(modContext=myModification2)
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/vm.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/core.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/charsets.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/graphics.jar
...
5: 0xE0FF6650 CLASSPATH
(modContext=myModification2)
/home/rumsbya
5: 0xE0FF6618 ROMCLASS: HelloWorld at 0xE010DC10.
Index 0 in classpath 0xE0FF6650
...
Figure 6-11 Content of the class cache

70 IBM Technology for Java Virtual Machine in IBM i5/OS

 Note we also created a second modification context called myModification2. Figure 6-11
shows that despite the number of modification contexts that may exist, if the class bytes are
identical there is only one ROMClass stored in the cache for each application that uses the
HelloWorld class, in this case at address 0xE010DC10.

Using modified bytecode with shared classes

Next we modify the bytes of the HelloWorld class on-the-fly to show that a different version of
the class is stored in the cache, in the separate area defined by the modification context. As
mentioned earlier, there are two broad ways of doing this, using a JVMTI agent or a
user-defined classloader to instrument the class before it is defined.

For this example, we use the HPROF JVMTI agent. HPROF is a simple demonstration JVMTI
agent supplied with IBM Technology for JVM. It can gather various data about an application,
such as time spent in each method and heap activity. HPROF uses bytecode instrumentation
depending on which options you specify at runtime. Not all options require the application to
be instrumented.

When you run Java with HPROF, an output file is created at the end of program execution.
This file is placed in the current working directory and is called java.hprof.txt. You can specify
additional command line argument file=<filename> to specify a different file name.

For our purposes we want to modify the bytecodes, therefore we use the cpu=times option
with HPROF. This inserts bytecodes into all method entry and exit points, therefore, clearly
the bytecodes of the HelloWorld class are going to differ from the version in the cache after it
is instrumented by HPROF. The command line and output are shown in Figure 6-12.

QSH Command Entry

> java -agentlib:hprof=cpu=times -Xshareclasses:name=myCache,modified=hprofcontext HelloWorld

Hello world!
Dumping CPU usage by sampling running threads ... done.
$
> java -Xshareclasses:name=myCache,printAllStats
...
1: 0xE0FFCD08 CLASSPATH
(modContext=myModification1)
/home/rumsbya
1: 0xE0FFCCD0 ROMCLASS: HelloWorld at 0xE010DC10.
Index 0 in classpath 0xE0FFCD08
...
7: 0xE0FF5E70 CLASSPATH
(modContext=hprofcontext)
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/vm.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/core.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/charsets.jar
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib/graphics.jar
...
7: 0xE0FF32EC CLASSPATH
(modContext=hprofcontext)
/home/rumsbya
7: 0xE0FF32B4 ROMCLASS: HelloWorld at 0xE022E4B0.
Index 0 in classpath 0xE0FF32EC
...
Figure 6-12 Using HPROF

Chapter 6. Optimizing performance with shared classes 71

 You can see that there are now two versions of the HelloWorld class in the cache, the original
version at 0xE010DC10 and the instrumented version at 0xE022E4B0.

72 IBM Technology for Java Virtual Machine in IBM i5/OS

 7

Chapter 7. Java Native Interface

This chapter discusses the Java Native Interface (JNI), the standard native programming
interface for Java to invoke native methods. This chapter provides an overview of the JNI and
discusses specific considerations for using the JNI in IBM Technology for JVM. This chapter
is not intended to be a comprehensive tutorial on how to use the JNI, however, 7.5, “Further
information about the JNI” on page 81 contains links to learning resources for using the JNI.

This chapter discusses the following topics:

򐂰 7.1, “Java Native Interface overview” on page 74
򐂰 7.2, “When must you use JNI?” on page 75
򐂰 7.3, “JNI considerations in IBM Technology for JVM” on page 76
򐂰 7.4, “Diagnostic messages and files” on page 81
򐂰 7.5, “Further information about the JNI” on page 81

© Copyright IBM Corp. 2007. All rights reserved. 73

7.1 Java Native Interface overview
The Java Native Interface (JNI) is the standard native programming interface for Java, which
is integrated into the Java Development Kit (JDK). JNI allows Java applications to interact with
applications and libraries that are written in other languages, such as C, C++, and RPG. In
addition, the Invocation application programming interface (API) defined as part of the JNI
allows you to embed a Java virtual machine (JVM) into your native application. This is useful
in situations where you want to use functionality of a Java application or class from your
native application.

JNI provides a standard way to interface with programs written in other languages. This
allows you to maintain a single version of your native method libraries on that platform.

Note: If an application uses the JNI, by virtue of relying on native code it ceases to be a
100% pure Java application. Therefore, if portability is important to you, investigate
alternatives to JNI such as the IBM Toolbox for Java classes.

The JNI has two main purposes:

򐂰 It specifies a way to write Java native methods. Java native methods invoke code written
in other languages native to the platform on which Java is running.
򐂰 It includes an Invocation API for embedding a JVM in native applications.

You can use JNI to write native methods to handle those situations when an application
cannot be written entirely in the Java programming language. For example, you might have to
use native methods and JNI in the following situations:
򐂰 The standard Java class library does not support the platform-dependent features that
your application requires.
򐂰 You have a library or application that is written in another programming language and you
want to make it accessible to Java applications.
򐂰 You want to implement a small portion of time-critical code in a lower-level programming
language, such as C, and have your Java application call these functions.

Programming with the JNI framework lets you use native methods to perform many
operations. You may use native methods to represent legacy applications or explicitly to solve
a problem that is best handled outside the Java programming environment. The JNI
framework lets your native method use Java objects in the same way that Java code uses
these objects.

The JNI serves as the glue between Java and native applications.

74 IBM Technology for Java Virtual Machine in IBM i5/OS

 Figure 7-1 shows how the JNI ties the native, or C, side of an application to the Java side.

Figure 7-1 JNI links Java code to native code

A native method can create Java objects, including arrays and strings, and then inspect and
use these objects to perform its tasks. A native method can also inspect and use objects that
are created by Java application code. A native method can even update Java objects that it
created or that were passed to it. These updated objects are available to the Java application.
Therefore, both the native language side and the Java side of an application can create,
update, and access Java objects and then share these objects between them.

Native methods can also call Java methods. Often you will already have developed a library of
Java methods. Your native method does not have to “re-invent the wheel” to perform
functionality that is already incorporated in existing Java methods. The native method, using
the JNI framework, can call the existing Java method, pass it the required parameters, and
get the results back when the method completes.

7.2 When must you use JNI?

C, C++, or RPG programmers who want to integrate with Java methods might want to start
programming in JNI. JNI protects your code from unknowns, such as the vendor-specific VM
extension that is used to integrate traditional language with Java. By conforming to the JNI
standard, a native library (C, C++, or RPG) is given the best option to run problem free in any
standard implementation of a JVM. The JNI specification does not impose any overhead or
restrictions on any JVM implementation, including object representation, garbage collection
scheme, and so on.

For the System i platform, several other alternatives to using the JNI are available including:
򐂰 IBM Toolbox for Java access classes such as Distributed Program Call and Data Queue
support
򐂰 The Extensible Program Call Markup Language (XPCML) support available with the IBM
Toolbox for Java

The IBM Toolbox for Java access classes and XPCML are easier to program than JNI. The
advantage of using JNI is that both the calling program and the called program run in the
same process (job) on the System i server, while some other methods, such as ProgramCall,
might use a new process (job). This makes JNI calls faster at startup time and less resource
intensive. However, using IBM Toolbox for Java access classes ensures the application
remains 100% pure Java for maximum portability.

Chapter 7. Java Native Interface 75

 For more information about the IBM Toolbox for Java and XPCML, refer to the System i
Information Center at:
http://publib.boulder.ibm.com/infocenter/iseries/v5r4/index.jsp?topic=/rzahh/rzahh
xmlmain.htm

7.3 JNI considerations in IBM Technology for JVM

Be aware of these considerations and caveats before using JNI in IBM Technology for JVM.
Also, there are some differences in requirements for native code used with JNI compared to
Classic JVM.

7.3.1 Teraspace storage model

If you want to use IBM Technology for Java and have programs that use native methods, you
must compile these programs with teraspace storage enabled. Because teraspace storage is
not enabled by default, it is likely that you have to recompile. This is necessary because the
Java object is in i5/OS PASE storage, which is mapped on top of teraspace storage, and a
teraspace storage pointer is returned. Also, the JNI functions, such as GetxxxArrayRegion,
have a parameter to a buffer where the data is placed. This pointer must point to teraspace
storage to enable the JNI function in i5/OS PASE to copy the data into this storage. If you
have not compiled your program with teraspace storage enabled and attempt to run the native
method with IBM Technology for Java, you are going to receive the escape message:
MCH4443 (Invalid storage model for target program LOADLIB)

When you recompile a native method, use the following options:

TERASPACE(*YES) STGMDL(*TERASPACE) DTAMDL(*LLP64)

For more information see InfoCenter at:

http://publib.boulder.ibm.com/infocenter/iseries/v5r4/index.jsp?topic=/rzaha/rzaha
teraspacemain.htm

7.3.2 Thread safety of i5/OS system functions

Not all i5/OS system functions are thread safe. Therefore, using the JNI support to invoke a C
function that calls a non-thread safe i5/OS system API might cause unpredictable results to
occur. The System API Reference lists the thread safe status of every system API. It is
recommended consulting this reference before using any system API through the JNI support
provided in IBM Technology for JVM. The System API Reference for the current V5R4 is
available at:
http://publib.boulder.ibm.com/infocenter/iseries/v5r4/topic/apifinder/finder.htm

7.3.3 64-bit PASE native methods

In V5R4 IBM Technology for Java Virtual Machine runs in a 32-bit PASE. If you have 64-bit
PASE native methods, you cannot use 32-bit IBM Technology for Java Virtual Machine to call
these methods and must use Classic JVM instead.

When 64-bit IBM Technology for JVM is available in i5/OS, you are going to be able to call
64-bit PASE native methods.

76 IBM Technology for Java Virtual Machine in IBM i5/OS

7.3.4 Adopted authority
Basically, a method can adopt the authority of the owner of the program being run rather than
the authority of the person running the program in order to accomplish some operation.
System i platform has provided the adopted authority feature for Java programs since the
JVM was part of the Trusted Computing Base in System Licensed Internal Code (SLIC).

As IBM moves to a Java implementation where the JVM is a user level program, Java can no
longer provide this feature. In order to achieve the same functionality, the most common
approach is to add an Integrated Language Environment (ILE) native method to the Java
program that adopts equivalent user profile authority and performs the required operation.

This section discusses two scenarios where it refers to some attributes of the native objects.
During runtime, an i5/OS program object identifies the adoption attributes of the invocation.
One of these attributes is Allow/Drop Adopted Authority. In the scenarios we specify the value
of this attribute in the parentheses. We also highlight a method which adopts authority.

Scenario 1
A Java method adopts authority immediately before calling a native method. See Figure 7-2.

Native method Y needing user profile

authority (allows)

Native method X (allows)

Java method J adopts user profile P

Java code...

Thread initiation

Figure 7-2 Scenario 1

In this example, Java method J is contained in a Java program which adopts user profile P
and directly calls native method X. Native method X calls ILE method Y which requires the
adopted authority. To run this Java program using IBM Technology for JVM, select one of the
following two alternatives.

Chapter 7. Java Native Interface 77

 Alternative 1
One way to preserve authority adoption is to segregate native method X into a new service
program. This new service program might then adopt user profile P which was previously
adopted by Java method J. This alternative is shown in Figure 7-3.

Native method Y needing user profile

authority (allows)

Native method X adopts user profile P

Java method J (which previously

adopted P)

Java code...

Thread initiation

Figure 7-3 Scenario 1: alternative 1

Alternative 2
Another way to preserve user profile adoption is to create an entirely new native method N
contained in a service program which adopts user profile P. This new method is called by Java
method J and calls native method X. Java method J requires to be changed to call N instead
of X, but native method X does not have to be changed or repackaged. Figure 7-4 shows
alternative 2.

Native method Y needing user profile

authority (allows)

Native method X (allows)

New native method N , which adopts

user profile P

Java method J (which previously

adopted P )

Java code...

Thread initiation

Figure 7-4 Scenario 1: alternative 2

78 IBM Technology for Java Virtual Machine in IBM i5/OS

Scenario 2
In the second scenario a Java method adopts authority and calls other Java methods before
calling a native method. See Figure 7-5.

Native method Y needing user profile

authority (allows)

Native method X (allows)

Java method J3 (allows)

Java method J2 (allows)

Java method J1 adopts user profile P

Java code...

Thread initiation

Figure 7-5 Scenario 2

In this example, a Java method J1 is contained in a Java program which adopts user profile P.
J1 calls Java method J2, which calls J3, which calls native method X. Native method X calls
ILE method Y which requires the adopted authority.

In order to preserve adopted authority in this case, a new native method N can be created.
This native method is contained in a service program which adopts user profile P. Native
method N would then use JNI to call Java method J2, which is unchanged. Java method J1
would require to be changed to call native method N instead of Java method J2.

Chapter 7. Java Native Interface 79

 Figure 7-6 shows the alternative solution.

Native method Y needing user profile

authority (allows)

Native method X (allows)

Java method J3 (allows)

Java method J2 (allows)

New native method N adopts user

profile P

Java method J1 (which previously

adopted P)

Java code...

Thread initiation

Figure 7-6 Alternative solution for scenario 2

7.3.5 Performance of native code

When making calls to ILE native code from Java applications through the JNI, you have to
consider that most calls have to cross the boundary between the PASE (in which IBM
Technology for JVM runs) and the SLIC. This adds significant overhead to the call.
Consequently, making frequent calls to small ILE native procedures or system programs is
the most unfavorable approach. The overhead of crossing the PASE-SLIC boundary is
comparable or greater than the execution time of the ILE code. Performance of JNI call to ILE
native code has been identified as a focus area for improvement.

On the other hand, calls to i5/OS PASE native methods from Java applications running under
IBM Technology for JVM, perform better than in the Classic JVM. This is because IBM
Technology for JVM runs in the same address space as i5/OS PASE native methods.

Attention: It is recommended that if a Java application must use JNI, then use i5/OS
PASE native methods wherever possible in preference to other native methods.
Performance of i5/OS PASE native methods is several orders of magnitude better than
other native methods.

80 IBM Technology for Java Virtual Machine in IBM i5/OS

7.4 Diagnostic messages and files
When ILE native methods encounter problems, you are going to see messages in the job log.
When IBM Technology for Java Virtual Machine or PASE native methods encounter
problems, they dump diagnostic files into the Integrated File System (IFS). There are several
types of these “core files”, including core.*.dmp, javacore.*.txt, Snap*.trc, and
heapdump.*.phd. The files range in size from tens of KB up to hundreds of MB. In most
cases, more severe problems produce larger files. The larger files can quickly and quietly
consume large amounts of IFS space. Despite the space these files consume, they are useful
for debugging purposes. When possible, you must preserve these files until the underlying
problem has been resolved.

Notice, however, that the presence of dump files does not indicate that you have a JNI
problem. These files can be generated for a wide variety of reasons.

7.5 Further information about the JNI

For more information about the JNI and its usage, the following resources are available:
򐂰 JNI 5.0 Specification (from Sun Microsystems, Inc.):
http://java.sun.com/j2se/1.5.0/docs/guide/jni/index.html
򐂰 The System i Information Center:
http://publib.boulder.ibm.com/infocenter/iseries/v5r4/index.jsp
򐂰 Chapter 8 of the IBM Developer Kit and Runtime Environment, Java 2 Technology Edition,
Version 5.0 Diagnostics Guide:
http://www-128.ibm.com/developerworks/java/jdk/diagnosis/

Chapter 7. Java Native Interface 81

82 IBM Technology for Java Virtual Machine in IBM i5/OS
 8

Chapter 8. Analyzing JVM behavior

This chapter helps you to identify and solve certain kinds problems that you might have while
running your application in IBM Technology for Java virtual machine (JVM). We take a
practical approach in this chapter. This chapter identifies five major problem areas and
demonstrates a process for finding the root cause of each type of problem. It demonstrates
tools provided with the IBM Support Assistant where possible and how you can use them in
the troubleshooting process. This chapter also presents a section on debugging Java
applications using WebSphere Development Studio Client for iSeries.

This chapter covers the following topics:

򐂰 8.1, “JVM analysis overview” on page 84
򐂰 8.2, “JVM does not start” on page 85
򐂰 8.3, “JVM crash” on page 86
򐂰 8.4, “Unresponsive application” on page 90
򐂰 8.5, “Investigating a suspected memory leak” on page 111
򐂰 8.6, “Performance issues” on page 127
򐂰 8.7, “Application debugging” on page 155

© Copyright IBM Corp. 2007. All rights reserved. 83

8.1 JVM analysis overview
This chapter focuses on a bottom-up approach to troubleshooting problems. This means
starting at the lowest level above the operating system, IBM Technology for JVM, the
components of the system up to and including the application code itself are analyzed in turn
to see if they are the cause of the problem.

There are many good resources for troubleshooting problems in a top-down manner. These
approaches assume there is some high-level symptom which is being observed or reported
by end users. For example, a Web-facing application might be timing out in a Web browser.

Troubleshooting in a top-down manner initially focuses on the application itself. If there is no

evidence of a problem with the application, the next stage is to troubleshoot the middleware
layer, for example WebSphere Application Server and associated components such as IBM
DB2®.

If you suspect the issue might be with WebSphere Application Server, it is particularly
recommended consulting the IBM Redbook WebSphere Application Server V6 Problem
Determination for Distributed Platforms, SG24-6798.

Using a bottom-up approach to troubleshooting is suited to cases where there is some visible
symptom occurring at a low level, typically this would be some anomaly in the JVM itself. For
example, the JVM job might disappear unexpectedly and no longer be visible through
WRKACTJOB. Or you might be seeing a memory leak in the form of heap exhaustion.
Therefore, the first stage is to make sure there is no issue with the JVM itself causing the
heap exhaustion.

If you find no evidence that the problem is with IBM Technology for JVM, then the next stage
would be to troubleshoot the middleware layer, for example WebSphere Application Server.
Logically, the next step after this would be to troubleshoot the application itself if no cause
was found for the problem in the middleware layer.

Using a bottom-up approach does not guarantee faster problem resolution. You might
eventually discover for instance that the application is making inefficient use of the heap by
preserving object references longer than is required or instantiating many unnecessary
objects. In this case, although the problem is with the application, the most visible symptom
was the heap running out of space at the JVM level.

In some situations you have to contact IBM Support in order to identify and fix the problem.
However, by following the recommendations in this chapter you can significantly reduce the
amount of time required for IBM to provide a fix or workaround.

The five major problem that this chapter covers are:

򐂰 JVM does not start.
򐂰 JVM crash.
򐂰 Application becomes unresponsive.
򐂰 Application performance issues.
򐂰 Memory leak.

84 IBM Technology for Java Virtual Machine in IBM i5/OS

 Figure 8-1 shows a diagram for using the bottom-up approach in troubleshooting IBM
Technology for JVM.

Does the No Does the No

application Run the application application start
start in the classic JVM. properly?
properly?

Yes Yes

Environment problem
Refer to chapter 4
Is the
application No Is the No
currently JVM job active?
running?

Yes Yes

Hang or deadlock JVM crash Application error

Does
Refer to chapter 8.4 Refer to chapter 8.3 Refer to chapter 8.7
No response
time
Increase over
time?

Yes

Application
performance Memory leak
Refer to chapter 8.6 Refer to chapter 8.5

Figure 8-1 Basic problem determination steps for IBM Technology for JVM

8.2 JVM does not start

If you have an application using IBM Technology for JVM and it does not start, you have to
isolate the problem into one of the following areas:
򐂰 Environment problem
򐂰 JVM problem
򐂰 Application problem

8.2.1 Quick fix

Before you start debugging a problem, it is recommended that you try the following tasks:
򐂰 Switch to Classic JVM
򐂰 Check group program temporary fix (PTF) level for Java

Use the Classic JVM

A good test is to run the application under the Classic JVM. Refer to 3.3, “Basic configuration”
on page 23 for information about configuring i5/OS to use the Classic JVM. Make sure you
use the correct Java version:
-Djava.version=1.5

Chapter 8. Analyzing JVM behavior 85

 If the JVM does not start with the Classic JVM, then the issue might be an environment
problem. However, if the application runs properly under Classic JVM, then there might be an
issue with IBM Technology for JVM. In either case, you are going to be closer to resolving the
issue because IBM Support has an understanding of whether it is likely a JVM or environment
issue and can apply more specific troubleshooting techniques.

Check PTF level

Check the fix level for Java by using the WRKPTFGRP CL command. Compare your level
with what is currently available at the following Web site:
http://www-912.ibm.com/s_dir/sline003.NSF/GroupPTFs?OpenView&view=GroupPTFs

If there is a newer fix level you must order and install it.

8.2.2 Environment problem

If your program executes when using the Classic JVM you could have a problem with the
setup of the IBM Technology for JVM environment. Try to check if the sample Hello program
runs? Refer to 3.2.3, “Verifying the installation” on page 22 for instructions.

If a simple program like Hello runs properly, you must investigate what properties and
environment variables are being set when you invoke your program.

Refer to Chapter 3, “New user guide” on page 19 for more information about setting
properties and environment variables for IBM Technology for JVM.

8.2.3 JVM problem

If the sample Hello program does not run, there is likely a problem with the installation of IBM
Technology for JVM or one of the prerequisite products such as i5/OS PASE. Refer to 3.2,
“Installing IBM Technology for JVM” on page 20 for more information.

8.2.4 Application problem

If the Hello program runs in both JVMs, but your application fails to start using either Classic
JVM or IBM Technology for JVM, the problem is likely specific to the application and must be
debugged as such.

8.3 JVM crash

Abnormal termination of a JVM is known as a crash. When a crash occurs, your application is
no longer running due to the JVM termination. A crash is characterized by the JVM job no
longer being visible through WRKACTJOB.

Depending on the environment where you run an application (production or development),

you might apply a different set of steps for resolving a crash. In production environment you
might want to restart an application as soon as possible and then analyze the cause of a
crash. In this case, you have to look at 8.3.2, “Quick fix” on page 88 as your first step.

More generally, the first step in analyzing a crash must be to check any log file and also look
for javacore files. In many cases (especially when it is not a true crash of the process) these
log files give an indication of what went wrong.

86 IBM Technology for Java Virtual Machine in IBM i5/OS

8.3.1 Causes of a crash
Crashes can occur for a variety of reasons, although most can be classified under three
broad categories:
򐂰 Environment problem
򐂰 JVM problem
򐂰 Application problem

Environment problems might include running out of memory, incorrectly setting environment
variables, and so on. Often problems in the environment involve some degree of user control,
while JVM problems might not be resolvable by the user beyond implementing a temporary
workaround.

JVM problems might include defects in the JVM itself or corrupt data in the JVM.

Application problems might include crashes due to poorly-written Java Native Interface (JNI)
or error-handling code.

Crashes might involve a certain degree of predictability. They might occur whenever a certain
set of conditions occur, or they might seemingly occur at random. For example, if a certain
transaction keeps crashing the JVM at the same point, then there is a good chance either the
data or methods associated with that transaction are causing the problem.

Crashes and native code

If your application loads any custom native libraries, for example a JVMTI agent to profile the
application, try running the application without loading these libraries if possible. If the JVM
does not crash, it suggests a problem with the native library.

You can check the Javadump file generated at the time of the crash (refer to “Locating a
Javadump file” on page 88) to confirm if the problem is in your native code. The file has the
following format:
javacore.DDDDDDDD.nnnnnn.nnnn.txt

Here DDDDDD is the date, and nnnnnn.nnnn is a time stamp.

It is recommended that you perform this before contacting IBM Support. The Javadump file
shows which library the crash occurred in, as shown in Example 8-1.

Example 8-1 Extract from Javadump showing faulty library caused a General Protection Fault
------------------------------------------------------------------------
0SECTION TITLE subcomponent dump routine
NULL ===============================
1TISIGINFO Dump Event "gpf" (00002000) received
1TIDATETIME Date: 2006/10/04 at 17:57:02
1TIFILENAME Javacore filename:
/home/rumsbya/jars/javacore.20061004.175702.9689.txt
NULL
------------------------------------------------------------------------
0SECTION GPINFO subcomponent dump routine
NULL ================================
2XHOSLEVEL OS Level : OS400 5.4
2XHCPUS Processors -
3XHCPUARCH Architecture : ppc
3XHNUMCPUS How Many : 2
NULL

Chapter 8. Analyzing JVM behavior 87

 1XHEXCPCODE J9Generic_Signal_Number: 0000000000000004
1XHEXCPCODE Signal_Number: 000000000000000B
1XHEXCPCODE Error_Value: 0000000000000000
1XHEXCPCODE Signal_Code: 0000000000000001
1XHEXCPCODE Handler1: 0000008000587708
1XHEXCPCODE Handler2: 000000800063AF00
NULL
1XHEXCPMODULE Module: /home/rumsbya/crashlib.so
...

In Example 8-1 because the crash occurred in a custom library which is not part of the JVM,
you must investigate any recent changes to the library which might have introduced a bug. If
you have a previous version that was known to work, you could revert to using that while
debugging the new library.

Locating a Javadump file

You can find this file using the following approach:
򐂰 The location specified by the IBM_JAVACOREDIR environment variable if set
򐂰 The current working directory of the JVM processes
򐂰 The location specified by the TMPDIR environment variable, if set
򐂰 The /tmp directory
򐂰 If the Javadump cannot be stored in any of the previous locations listed, then it is put in to
STDERR

8.3.2 Quick fix

If the JVM crashes, it is important to get the application up and running again to reduce the
impact on the business and end users. It might be possible to workaround the crash using
one of a variety of methods. Doing so is in your best interests because it reduces the impact
of the problem to end users and your business. A temporary workaround is also going to “buy
some time” to allow you to work with IBM Support on a permanent resolution.

If the crash is occurring in a test environment rather than production, then the importance of a
quick fix might be lessened. However, the approaches that follow are still useful in narrowing
the scope of where the issue is occurring and allows more specific troubleshooting steps to
be followed. This particularly applies in cases where IBM Support are involved.

Use the Classic JVM

A good test is to run the application under the Classic JVM (refer to 3.3, “Basic configuration”
on page 23 for information about configuring i5/OS to use the Classic JVM).

If the Classic JVM also crashes, then the issue might be an environment problem. However if
the application runs properly under Classic JVM, then there might be an issue with IBM
Technology for JVM. In either case, you are closer to resolving the issue because IBM
Support has an understanding of whether it is likely a JVM or environment issue and can
apply more specific troubleshooting techniques.

Disable the just-in-time compiler

The Java runtime environment can be viewed as two basic components, which are closely
bound:
򐂰 The JVM itself
򐂰 The just-in-time (JIT) compiler

88 IBM Technology for Java Virtual Machine in IBM i5/OS

 The JIT compiler generates native code from Java bytecodes. The native code is then
executed directly, which improves performance. The JIT compiler is enabled by default. For
further information about the JIT compiler, refer to “Just-in-time compiler” on page 6.

Although considered very rare, a failure in the JIT compiler might cause the JVM to crash. For
this reason, if the JVM fails, it is useful to rule out the JIT compiler as a possible cause early
in troubleshooting.

Therefore, it is recommended that you run the application with the JIT compiler disabled. To
disable the JIT compiler, start the application with the -Xint option specified on the command
line. For example, issuing the following command from a Qshell prompt causes the
HelloWorld application to be run in purely interpreted mode:
java -Xint HelloWorld

If the application then runs properly, the issue mighty be related to the JIT compiler. In this
case it is recommended that you contact IBM Support to report the failure. You can refer to
the IBM Developer Kit and Runtime Environment, Java 2 Technology Edition, Version 5.0
Diagnostics Guide for instructions on how to identify the specific method causing the JIT
compiler failure. The IBM Developer Kit and Runtime Environment, Java 2 Technology
Edition, Version 5.0 Diagnostics Guide is available on the Web at:
http://www-128.ibm.com/developerworks/java/jdk/diagnosis/

Identifying the failing method means you can selectively disable compilation of just that
method as opposed to all methods. The benefit of applying a more granular approach and
disabling JIT compilation of just the failing method is that the vast majority of methods can still
be compiled into native code. As a result, you only lose a negligible fraction of JVM
performance in comparison to using interpreted mode only.

For example, issuing the following command from a Qshell prompt prevents JIT compilation
of just the Math.max() method when the HelloWorld application runs:
java -Xjit:exclude={java/lang/Math.max(II)I} HelloWorld

This method, instead, is executed in interpreted mode, bypassing the JIT compiler. All other
methods are eligible for JIT compilation.

8.3.3 If application runs properly under Classic JVM

There might be an issue with IBM Technology for JVM. You can contact IBM Support to
report the issue and continue to use the Classic JVM as a temporary workaround. You might
also wish to do some basic troubleshooting yourself, which might lead to a workaround under
IBM Technology for JVM.

Heap differences
One case where the application might run properly under the Classic JVM, but not under IBM
Technology for JVM, is where the JVM runs out of memory. Because of the approximate
3 GB heap size limit with the 32-bit IBM Technology for JVM, the application might not fit
within the 32-bit address space. In this case you must continue to run the application under
Classic JVM until such time that the 64-bit IBM Technology for JVM becomes available.

If you are experiencing crashes in IBM Technology for JVM but not Classic JVM, you must
increase the maximum heap size as much as possible. If the application still uses all the heap
and crashes, then it might be that the application’s heap requirements are too large for IBM
Technology for JVM.

Chapter 8. Analyzing JVM behavior 89

 The other possibility is a memory leak. Due to the smaller maximum heap size, the effect of a
leak is more pronounced in IBM Technology for JVM. With the Classic JVM an out of memory
condition could take longer to occur or might not occur at all, because of the unlimited
maximum heap size. Refer to 8.5, “Investigating a suspected memory leak” on page 111 for
information about troubleshooting a memory leak.

8.3.4 If the Classic JVM also crashes

If the Classic JVM also crashes, then there might be an environment issue. The application
might be exhausting the heap, which could occur for several reasons:
򐂰 Memory is leaking
򐂰 The maximum heap size has been set too small for the application’s requirements

A memory leak might be a more likely cause in this scenario where both JVMs crash,
therefore, refer to in 8.5, “Investigating a suspected memory leak” on page 111 for guidance.

Try specifying a larger maximum heap size using the -Xmx command line parameter. For
example, the following uses a maximum 3 GB heap size instead of the default 2 GB under
IBM Technology for JVM:
java -Xmx3g HelloWorld

However, if the application crashes with an unlimited heap under Classic JVM, specifying a
larger maximum heap size under IBM Technology for JVM is unlikely to alleviate the problem.

8.3.5 Identifying the cause of a crash

Reviewing the Javadump file (javacore*.txt) created during a crash indicates in which library
the crash occurred. Refer to “Crashes and native code” on page 87 for an example.

It is straightforward from the library name and path whether the crash occurred in a JVM
library or a custom library that might be used by an application. JVM libraries typically have a
libj9 prefix and are located in the following directory by default:
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/lib

As explained earlier, if the crash is in a custom library, this has to be debugged by the
provider of the custom library. Crashes in JVM libraries must be reported to IBM Support for
further investigation.

Regardless of whether the crash occurs in a JVM or custom library, using debugging tools
such as those provided with WebSphere Development Studio Client for iSeries can help
isolate the area of application code which triggers the crash. This assists with the problem
isolation and determination effort. Refer to 8.7, “Application debugging” on page 155 for more
information about using WebSphere Development Studio Client for iSeries to do application
debugging.

8.4 Unresponsive application

This section outlines possible causes of an unresponsive application and how to troubleshoot
a hang. We present the ThreadAnalyzer tool as an aid to identifying deadlocks and analyzing
threads, both in WebSphere Application Server environments and stand-alone Java
applications. We also describe how WebSphere Development Studio Client for iSeries can be
used to investigate a suspected deadlock scenario.

90 IBM Technology for Java Virtual Machine in IBM i5/OS

8.4.1 Overview of a hang
An application that has stopped responding, but whose JVM is still running, is considered to
have hung. A hang is differentiated from a crash in that a hung JVM is displayed in the
WRKACTJOB output.

A hang can occur due to:

򐂰 Deadlock
򐂰 Infinite loop (spin)/CPU ‘hogging’
򐂰 Timeout
򐂰 Blocked on resource
򐂰 Memory constraints

8.4.2 Hangs in applications running on WebSphere Application Server

If the application is running on WebSphere Application Server rather than as a ‘stand-alone’
Java application, then there are higher-level WebSphere Application Server-specific
troubleshooting tools and techniques that you might wish to try first. The troubleshooting
steps presented in this section are focused toward stand-alone Java applications and do not
take account of advanced Reliability, Availability, and Serviceability (RAS) features built-in to
WebSphere Application Server. For example, WebSphere Application Server includes a hang
detection option which is enabled by default. This monitors managed threads and reports in
the log when a thread has been executing too long within a unit of work, because such a
thread may have hung.

If you plan to contact IBM Support about a hang issue with WebSphere Application Server,
IBM Support Assistant can be used to collect the “MustGather” data required by IBM Support.
Refer to “Submitting a problem report to IBM” on page 187 for more information.

8.4.3 Quick fix

Although restarting the JVM alleviates the problem, if the application runs long enough, the
hang conditions are likely to occur again if no corrective action is taken. Therefore, if a hang
occurs, it is recommended that you take some time to review this chapter and try some of the
troubleshooting steps that are appropriate to your particular situation.

It is rare that a hang is due to an underlying JVM error. In addition to a JVM error, a hang can
occur due to a poorly-written application or problem with the system environment. For
example, if threads are not properly synchronized, deadlock can arise if resources are shared
between multiple threads and contention occurs.

It is likely that a hung JVM has to be restarted because it is unlikely to make any progress in
such a degraded state. If the hang occurs in a test environment, however, you might want to
wait to determine if any progress is made if the application is left long enough.

Running the application under Classic JVM is a valid troubleshooting step. If the hang occurs
in Classic JVM, then the application might be at fault. If no hang occurs, then there might be
an issue with IBM Technology for JVM.

In situations where the hang occurs only under IBM Technology for JVM, it is recommended
that you run the application with the JIT compiler disabled. Refer to “Disable the just-in-time
compiler” on page 88 for information about how to disable the JIT compiler.

Chapter 8. Analyzing JVM behavior 91

 If the hang still occurs with the JIT compiler disabled, you have at least ruled out the JIT
compiler as a possible cause. At this stage, because the Classic JVM and IBM Technology
for JVM are exhibiting different behavior, you might want to engage IBM Support which is
going to supply more specific troubleshooting steps.

For situations where the hang occurs under both JVMs, you must investigate the hang further
to determine if it is an issue with the application or the environment. If the application tends to
run for a relatively long period before hanging, a temporary workaround to “buy some time”
for further investigation might be to schedule a periodic restart of the application. If you can
identify with some degree of confidence how often or how long after startup the application
hangs, you can schedule a restart before the expected window where the hang could occur,
to reset the application environment and therefore reduce the likelihood of the hang
occurring.

Under any circumstances you must take a javadump (refer to “Generating the Javadump” on
page 93) before ending or restarting the JVM. Even if you do not intend to look at it, the
javacore can be very helpful to IBM support.

8.4.4 Troubleshooting a hang

This section demonstrates methods and tools which you can use to troubleshoot a hang.
Typically, you would refer to this section if you have a hang that occurs under both Classic
JVM and IBM Technology for JVM, in which case the source of the hang is more likely to be
related to the application or environment. We present an approach for each of the major
causes of hangs.

In which order you try the troubleshooting steps depends on the symptoms you are seeing
and your preference for using system tools or the ThreadAnalyzer tool with IBM Support
Assistant. It is worth mentioning that the system tools do not readily help you detect hangs
due to deadlock. In this case it is recommended that you capture a Javadump which you can
use as input to ThreadAnalyzer for deadlock detection which is reported in the
ThreadAnalyzer summary section.

This section also describes how you can use WebSphere Development Studio Client for
iSeries to investigate a suspected deadlock scenario.

Troubleshooting deadlock with ThreadAnalyzer

The ThreadAnalyzer tool available via the IBM Support Assistant (refer to Appendix B, “IBM
Support Assistant” on page 181) can detect whether any deadlocks are occurring.
ThreadAnalyzer is designed primarily to analyze Javadump files (also known as thread
dumps) from WebSphere Application Server. Because the Javadump is created at the JVM
level, however, you can also use it to analyze stand-alone applications.

You can analyze thread usage at several different levels, starting with a high-level graphical
view, and drilling down to a detailed tally of individual threads. If any deadlocks exist in the
thread dump, ThreadAnalyzer detects and reports them. It is straightforward using
ThreadAnalyzer to establish if deadlock is the cause of a hang, which is very useful in our
“dining philosophers” example.

Example stand-alone deadlocked application

The example application is a classic “dining philosophers” scenario leading to circular
deadlock. Such a deadlock involves multiple threads where each requires a resource held by
another in order to progress.

92 IBM Technology for Java Virtual Machine in IBM i5/OS

The basic premise of the “dining philosophers” scenario is that a group of philosophers go to
dinner. At the restaurant, there are only a limited number of knives and forks with which to
eat. This means not enough knives and forks for everyone to eat at the same time. A
philosopher is either eating, waiting for the knife/fork, or thinking between mouthfuls. A
philosopher eating at this particular restaurant is restricted as follows:
򐂰 They can only eat if they have both the knife and fork.
򐂰 They can only pick up either the knife or a fork at any time, but not both.
򐂰 After a philosopher picks up either a knife or fork, they cannot put it down until they have
eaten something.

The philosophers are independent people, therefore, each of them eats or waits or thinks for
different lengths of time. Because of this, inevitably the situation eventually arises that one
philosopher has the fork, while another has the knife. Both are waiting for the other to release
the utensil they require. However, because of the last restriction that neither might put their
utensil down until they have eaten, the philosophers cannot make any progress and are
therefore deadlocked.

The deadlock in this example is usually resolved by lifting one of the restrictions, such as the
restriction that a philosopher might not put the knife/fork down until they have eaten
something. The philosopher instead waits for a period of time, and if they do not get the other
utensil, then they put down the one they currently hold without having eaten. Such a move
prevents deadlock from occurring in future. In an application this would be implemented by a
thread yielding the resource it currently holds, after waiting unsuccessfully for the other
resource it requires, to become available. The thread would then try again after a random
period of time.

Generating the Javadump

A Javadump file is required as input to ThreadAnalyzer. This can be generated
non-destructively by giving the following command in a Qshell session. You have to open a
new Qshell session because the session running the JVM is not going to be responsive to
input:
kill -QUIT process ID

Important: The JVM continues running after the kill command is issued. However,
because the application cannot progress in a hung state, you can eventually terminate the
JVM job manually through WRKACTJOB in order to restart the application.

The process ID is that of the hung JVM. It can be found at the top of the JVM’s job log. To
display the job log, enter WRKACTJOB. Put an Option “5-Work with” next to the hung JVM, then
choose Option “10-Display job log...”

In our case the process ID of the hung JVM is 9793 as shown in the job log in Figure 8-2.

We run the following command in a Qshell session to generate the Javadump required by
ThreadAnalyzer:
kill -QUIT 9793

Refresh the job log by pressing F5. You can see messages posted in the job log as the
Javadump is created, also shown in Figure 8-2.

The name and location of the Javadump file is given in the job log so that you can find it
easily. In our case the Javadump file is as follows:
/home/rumsbya/javacore.20061004.140450.9607.txt

Chapter 8. Analyzing JVM behavior 93

 See Figure 8-2.

Display Job Log

System: RCHAS60
Job . . : QP0ZSPWP User . . : RUMSBYA Number . . . : 210932

Java Virtual Machine is IBM Technology for Java. PID(9793)

JVM properties were loaded from a properties file.
Public write authority on "bughunt.jar".
JVMDUMP006I Processing Dump Event "user", detail "" - Please Wait.
JVMDUMP007I JVM Requesting Java Dump using
'/home/rumsbya/javacore.20061004.140450.9607.txt'
JVMDUMP010I Java Dump written to
/home/rumsbya/javacore.20061004.140450.9607.txt
JVMDUMP013I Processed Dump Event "user", detail "".

Bottom
Press Enter to continue.

F3=Exit F5=Refresh F10=Display detailed messages F12=Cancel

F16=Job menu F24=More keys
Figure 8-2 The process ID of a hung JVM is found in the job log

Analyzing the Javadump with ThreadAnalyzer

The next step is to launch ThreadAnalyzer from IBM Support Assistant and determine if there
are any deadlocks:
1. Ensure IBM Support Assistant is running. Refer to Appendix B, “IBM Support Assistant”
on page 181 for more information about installing and configuring IBM Support Assistant.
2. Click Tools on the IBM Support Assistant Welcome Page.
3. Click WebSphere Application Server 6.1 under Products.
4. Click ThreadAnalyzer as shown in Figure 8-3. A separate window opens with a message
Launching ThreadAnalyzer Tool....

Figure 8-3 ThreadAnalyzer is accessed from the Tools menu in IBM Support Assistant

94 IBM Technology for Java Virtual Machine in IBM i5/OS

 After ThreadAnalyzer starts, you have to import the Javadump file for analysis:
5. Either copy the Javadump file from the i5/OS to the PC where you are running
ThreadAnalyzer, or map a drive letter to the i5/OS filesystem. We copied the Javadump
file locally.
6. Select Open Existing File from the Thread Dumps menu. Select the Javadump file
generated for this hang, as shown in Figure 8-4.

Figure 8-4 Select the Javadump to process with ThreadAnalyzer

7. Click Open to import the Javadump into ThreadAnalyzer

8. ThreadAnalyzer then processes the Javadump. Click OK on the Open Existing File(s)
dialog when ThreadAnalyzer has completed processing.
9. Expand the root entry for the Javadump just processed (in this case there is only one
entry) then click Summary.

Chapter 8. Analyzing JVM behavior 95

 This displays the summary information, as shown in Figure 8-5.

Figure 8-5 ThreadAnalyzer summary shows the deadlock

From the summary information you can see a deadlock has been detected by
ThreadAnalyzer. Notice the “bang” icon next to the Javadump file name which gives visual
indication deadlock has been detected.
If you click Overall monitor analysis, you can view the details of the deadlock, as shown
in Figure 8-6.

Figure 8-6 Overall monitor analysis section of ThreadAnalyzer gives more detail about the deadlock

96 IBM Technology for Java Virtual Machine in IBM i5/OS

 Continuing the dining philosophers analogy, the results from ThreadAnalyzer indicate three
philosophers have picked up a utensil each, but each requires another utensil (held by
another of the two other philosophers) in order to continue. Because of the third restriction
(that after a knife/fork has been picked up, it cannot be put down until the philosopher has
eaten) none of the philosophers can make any progress and therefore a circular deadlock
has been reached.

In this case because the deadlocked threads are application threads, the application would
have to be debugged further to determine the cause. Refer to “Analyzing deadlocks in
WebSphere Development Studio Client for iSeries” on page 104 for guidance on
troubleshooting an application using WebSphere Development Studio Client for iSeries.

Using ThreadAnalyzer, you can view the call stacks of application threads. This is useful
when troubleshooting a hang due to deadlock. You can establish which method the thread
was executing at the time of the hang, narrowing the scope of where the problem could lie.
This also clarifies if the hang is likely an application issue or (less likely) a problem with one of
the Java shared class libraries.

Use ThreadAnalyzer to look at the Javadump from the dining philosopher’s example again to
see which methods were being executed when the deadlock occurred.
1. If ThreadAnalyzer is not currently running, start ThreadAnalyzer and load the Javadump
from the hang into ThreadAnalyzer. Refer to “Analyzing the Javadump with
ThreadAnalyzer” on page 94 for more information.
2. Select the Javadump file which was generated from the hang and click Open.
3. Expand the Overall thread analysis subfolder of the report:
a. Expand the root entry for this Javadump
b. Expand the Analysis subfolder
c. Expand the Overall thread analysis subfolder. You can see a summary of the
methods and how many times each method is found to be at the top of a thread’s
stack, as shown in Figure 8-7.

Figure 8-7 The overall thread analysis shows the occurrences of each method among the threads

Chapter 8. Analyzing JVM behavior 97

 Note: You might notice ThreadAnalyzer occasionally reports a thread has an empty stack.
This can occur if the Javadump supplied to ThreadAnalyzer does not contain the stack
trace for a thread. This is not an issue with ThreadAnalyzer.

Investigate the stack trace of the hung threads. Click the Eatery DINER_0 thread. The thread
information, including the call stack trace, is displayed, as shown in Figure 8-8.

Figure 8-8 Viewing the stack trace of a deadlocked thread in ThreadAnalyzer

Notice the stack trace reported in the thread information. This shows that the run() method
was active when the hang occurred. Therefore, for further troubleshooting you must
investigate the code in the run() method for logic errors that could lead to a hang. 8.6,
“Performance issues” on page 127 shows you how to use WebSphere Development Studio
Client for iSeries to profile the application.

Example WebSphere Application Server-based application

The ThreadAnalyzer tool is primarily designed for analyzing thread dumps (Javadumps)
captured in WebSphere Application Server environments, where there can be hundreds of
threads in an application. For such applications, manual analysis of thread dumps is
time-consuming and unreliable in pinpointing the cause of a performance or hang-related
problem.

Note: This section applies only to WebSphere Application Server V6.1 and later, because
support for IBM Technology for JVM was introduced in this version.

Deadlock situations are reported by ThreadAnalyzer, similar to stand-alone applications, as

described in “Analyzing the Javadump with ThreadAnalyzer” on page 94.

98 IBM Technology for Java Virtual Machine in IBM i5/OS

The remainder of this section is focused on issues where an application running under
WebSphere Application Server exhibits poor performance or hangs without a deadlock.

ThreadAnalyzer includes a custom filter feature that allows you to reduce the amount of
thread data that is displayed in the thread analysis view. You can also work with multiple
thread dumps simultaneously in a comparative analysis.

We used ThreadAnalyzer to investigate a performance issue in a WebSphere Application

Server environment. A Web application was giving very poor performance under WebSphere
Application Server. We captured several thread dumps during the application lifetime so that
we could use the ThreadAnalyzer’s comparative analysis feature to determine which threads
were making progress and what methods they were calling.

Note: Using ThreadAnalyzer to troubleshoot WebSphere Application Server environments

requires you to understand the architecture of the application and be able to distinguish
between core WebSphere Application Server methods and methods which belong to the
application.

The procedure used to troubleshoot this particular application is as follows:

1. Start the application and recreate the conditions which cause poor performance or monitor
for signs of poor performance.
2. Capture multiple thread dumps (Javadumps) when the problem surfaces.
Refer to “Generating the Javadump” on page 93 for information about how to capture a
Javadump. Because the application we are analyzing is running under WebSphere
Application Server, the process ID of the WebSphere Application Server JVM has to be
passed to the kill -QUIT command.
You can distinguish the WebSphere Application Server JVM from others that might be
running, because it will be running in the QWAS61 subsystem.
There is no “magic number” of thread dumps to capture, it varies by application and how
soon the problem surfaces.
3. Either copy the Javadump files from the i5/OS to the PC where you are running
ThreadAnalyzer, or map a drive letter to the i5/OS filesystem. We copied the Javadump
files locally.
4. Launch ThreadAnalyzer from the IBM Support Assistant. Refer to “Analyzing the
Javadump with ThreadAnalyzer” on page 94 for information about how to start the
ThreadAnalyzer.
5. Open the thread dumps to be analyzed. Because we are comparing multiple thread
dumps, use Ctrl-click to select multiple thread dumps to analyze. Refer to “Analyzing the
Javadump with ThreadAnalyzer” on page 94 for information about how to open thread
dumps.
6. Expand Multiple Dump Analysis under the Project folder, then select Custom Filter.
The resulting view sorts each method by the total number of threads which have that
method on the top of their stack, as shown in Figure 8-9.

Chapter 8. Analyzing JVM behavior 99

 See Figure 8-9.

Figure 8-9 Multiple Dump Analysis aggregates data from multiple Javadump files

The number of occurrences of threads with each method at the top of the stack, is
reported in a separate column for each Javadump also. Comparing these figures gives an
easy indication if the application has made any progress in the time between capturing
each Javadump. For larger applications, there can be hundreds of threads, and
correspondingly, hundreds of methods listed in this view.
Many of the methods are executed by core WebSphere Application Server threads,
because the application we are analyzing runs in a WebSphere Application Server
environment. It is easier to work with just the application threads that are involved in the
problem and discount the WebSphere Application Server threads.
This particular application was composed mostly of Servlets, therefore we want to focus
on the threads that run the servlet code, that is Servlet handler threads.

100 IBM Technology for Java Virtual Machine in IBM i5/OS

 7. In the Filter by Thread Name field enter Servlet then click Apply. The number of methods
displayed is reduced, as shown in Figure 8-10.

Figure 8-10 ThreadAnalyzer’s Custom Filter feature lets you focus on methods related to the application

You then have to analyze the methods further. Start with the top method listed in the
summary because it was the most common method being executed when the thread
dump was captured.
8. Double-click the top method, in this case java.net.SocketInputStream.socketRead().
Another view opens showing a tree structure under which the threads that have this
method at the top of their stack, are displayed.

Chapter 8. Analyzing JVM behavior 101

 9. Click the first thread dump file name to show the threads that have this method at the top
of the stack. Notice the thread’s call stack is displayed on the right, as shown in
Figure 8-11.

Figure 8-11 Reviewing a thread’s call stack in ThreadAnalyzer

The next stage is to analyze the thread call stacks that have this method at the top. You
are looking for a clue or common application methods which opens in many threads’
stacks. If you find such a method, it warrants further investigation.
10.We did this by pressing the down arrow key and watching the stack in the right pane, to
see if and how it changed between each thread. We noticed that the method writeToLog()
was displayed in every thread’s call stack.
From the package name, we knew this was not a method from one of the classes supplied
in the standard Java Class Libraries. Therefore, we decided to investigate this method
further.

102 IBM Technology for Java Virtual Machine in IBM i5/OS

 11.We further filter the thread data displayed by specifying writeToLog in the Filter by a
method in the stack trace field, then clicking Apply, as shown in Figure 8-12.

Figure 8-12 Finding occurrences of a method using ThreadAnalyzer’s custom filter feature

From Figure 8-12, you can see that it is only threads with the
java.net.SocketInputStream.socketRead() method at the top of the stack, which also have
called the writeToLog() method.
Remember that the java.net.SocketInputStream.socketRead() method was the most
common “top of stack” method in the application. The implication is that most of the
threads also spend time in the writeToLog() method.
12.This insight prompts you to ask the following questions:
– What is the function of the writeToLog() method?
– Why is it called by so many threads?
– Why does it take so long to return?
13.These are questions that you resolve by going to the source code for the writeToLog()
method.
After reviewing the source code, you discover this method is used for logging to a custom
log file. The method makes a synchronous call to a remote JVM to do the logging. The
inefficient operation of this method, combined with the sheer number of times it is called,
causes noticeable perturbation to the application.
The issue is resolved in this instance by rewriting the writeToLog() method to use
asynchronous logging by the current JVM rather than using expensive remote,
synchronous calls.

Chapter 8. Analyzing JVM behavior 103

 Analyzing deadlocks in WebSphere Development Studio Client for
iSeries
A thread deadlock occurs when two independent threads of control are blocked, each waiting
for the other to take some action. Two threads waiting for each other can bring an application,
or one section of the application, to a complete halt.

A thread contention, or race conditions, occurs when a thread is waiting for a lock or resource
that another thread holds. Programmers often add synchronization mechanisms to avoid
these contentions, but it is possible that the synchronization itself can lead to deadlocks, if
done incorrectly. In this section you see how to identify a deadlock situation between threads.

Restriction: At the time of this writing, WebSphere Development Studio Client for iSeries
version 6.0.1 has not supported J9 for remote debugging. However the approach
presented in this section is applicable to IBM Technology for JVM when this is supported
for remote debugging.

Perform the following steps:

1. Start WebSphere Development Studio Client for iSeries Advanced Edition on your
workstation.
2. Open the Java perspective.
3. Locate the Java application project and source files for your application.
4. Right-click your mouse on the class (which must have a main() method) and select
Profile → Java Application.
5. At the properties window select the Profiling tab.
6. Select the Thread Analysis option and click OK as shown in Figure 8-13.

Figure 8-13 Thread analysis selection

104 IBM Technology for Java Virtual Machine in IBM i5/OS

 7. If you do not have the profiling and logging perspective open you see a “confirm
perspective switch?” window open. If so, click Yes.
8. You must be in the profiling and logging perspective and see your application status as
monitoring. If the console does not open, go to the toolbar and select Window → Show
View → Console. This is going to display any output data.
9. You have to run the application to completion.
10.After the application completes you must see your application status as terminated.
Expand the application link and you must see a Thread Analysis option. Right mouse click
the Thread Analysis option and select Open With → Thread view.
11.The thread view is displayed, as shown in Figure 8-14. If there are thread deadlocks
detected, then they are indicated.

Figure 8-14 Default thread view example

The vertical arrows between threads are of interest. An arrow indicates that one thread
(the thread in which the arrow originates) is waiting for another thread (the thread to which
the arrow is pointing) to release a lock. A double arrow indicates that two threads are in a
deadlocked state, both waiting for the other to release a lock.
12.The initial view may be difficult to work with. Click Switch to Compressed Time button to
view a more user friendly representation of the thread activity, as shown in Figure 8-15.

Figure 8-15 Alternative thread view

Chapter 8. Analyzing JVM behavior 105

 Soon after the program starts, four philo#x threads are created, one after the other. Each
thread runs for a short while and then makes a request for a lock. The requests are not
successful. When the first three lock requests fail, the threads enter a Waiting for Lock
state. When the lock request for the fourth thread fails, deadlock occurs, and the display
indicates that all four threads are in a Deadlocked state.
Notice the vertical arrows between the threads. An arrow indicates that one thread (where
the arrow originates) is waiting for another thread (where the arrow points) to release a
lock. A double arrow indicates that two threads are in a deadlocked state, both waiting for
the other to release a lock.
In this example the philo#x threads are waiting for a lock from other philo#x threads that
are also waiting for a lock. There is now a thread deadlock and the program cannot
continue.
To resolve this issue you have to find those method calls and objects that are involved.
You can use the UML2 Sequence Diagram (the Object Interactions and Thread
Interactions views) and the Call Stack View, in addition to the Thread View to assist you.
13.In the Thread View, find the first thread that enters a “Deadlock” state. Pause your cursor
over the “Deadlock” segment. You must see the lock (Fork.<id number>) and the locking
thread (Locking Thread.<name>). For example, as follows:
Fork.8008
Locking Thread.philo#1
See Figure 8-16.

Figure 8-16 Identify the thread lock source

106 IBM Technology for Java Virtual Machine in IBM i5/OS

 14.Right-click the Profiling resource for the run and select Open With → UML2 Object
Interactions, as shown in Figure 8-17.

Figure 8-17 Opening the UML2 Object Interactions view

15.The UML2 Trace Interactions view opens, showing the object Interactions.
16.Scroll across the view to find the appropriate Fork.<id number> icon and select it, as
shown in Figure 8-18.
17.Scroll down the view to find the Locking Thread.<thread id> interaction.

Figure 8-18 Locate the object interaction

Chapter 8. Analyzing JVM behavior 107

 18.Click the method (getName() in Figure 8-18) and the arrow must be highlighted as shown
in Figure 8-19. You must also see a cursor in the thread view for this point in time.

Figure 8-19 View thread status at program call time

19.In this example you can see that thread philo#1 has started to run and has seized the Fork
object (Fork:8008 in this example). In the Thread view you can see that there are no
threads waiting for locks or any deadlocks.
20.In the UML2 trace interactions view scroll down to find the request for the same resource,
that originated in a different thread (the getName() method in this example).
21.Click the getName() method and the arrow must be highlighted as shown in Figure 8-20.
You must also see a cursor in the thread view for this point in time.
In this example you can see that thread philo#3 is starting to run and several other threads
have started to run but then stopped to wait for a lock. This indicates a deadlock is going
to occur shortly. The next task is to determine the method that is causing the problem.

108 IBM Technology for Java Virtual Machine in IBM i5/OS

 See Figure 8-20.

Figure 8-20 Thread locking example

22.Right-click the Profiling resource for the run and select Open With → UML2 Thread
Interactions.
23.In the Thread View, click the menu drop-down button and select Open Call Stack View as
shown in Figure 8-21.

Figure 8-21 View the thread interactions and call stack

Chapter 8. Analyzing JVM behavior 109

 24.Locate the first deadlocked thread graph in the thread view.
25.Scroll through the information for the thread and double-click the last method that the
thread ran successfully. The call stack view displays all the calls on the stack at that time
as shown in Figure 8-22.

Figure 8-22 View the call stack for the active thread

26.At this time you must analyze the source code where the deadlock has occurred (based
on the call stack). Select the Open source option to open source code as shown in
Figure 8-23.

Figure 8-23 Open the source for the method identified with the deadlock

Troubleshooting memory constraints

You saw that the JVM might crash if there is insufficient memory. Before such a situation
occurs, the application performance starts to degrade as available memory becomes more
scarce. Classic JVM memory might start being paged out to disk and thrashing might occur
whereby the system spends most of its time swapping pages between memory and disk. This
leaves less processing time available for the application to progress. Such a situation shows
itself through worsening response times and gives the impression that the application has
hung.

With IBM Technology for JVM, however, depending on how much memory is installed, the
JVM is likely to run out of memory before i5/OS has to start paging. This is because IBM
Technology for JVM can use only approximately 3 GB of memory, which might be
containable in main memory without the requirement to page. Therefore, you cannot
necessarily rely on paging to occur to let you know that the JVM is having a memory issue
when using IBM Technology for JVM.

110 IBM Technology for Java Virtual Machine in IBM i5/OS

 You can quickly check the WRKSYSSTS output. If you are seeing page faults in the pool
used by the JVM, then there is clearly a memory usage problem. If the cause of the hang has
not been found by this stage, you must check for a memory leak by following the guidance
provided in 8.5, “Investigating a suspected memory leak” on page 111.

8.5 Investigating a suspected memory leak

Java memory leak issues are probably the most frustrating and time consuming problems to
troubleshoot. If you suspect a memory leak in your Java application you are likely
experiencing one or more of the following symptoms:
򐂰 Abnormal end/crash due to an OutOfMemory condition
򐂰 High page faults (observed via WRKSYSSTS or performance data)
򐂰 Slow application response times

This section describes how to approach the solving of such problems. This sections begins
by offering some simple suggestions and moves toward some more advanced analysis tools
and techniques.

8.5.1 Quick fix

As already mentioned that memory leaks are difficult issues to resolve, the chances of finding
a “quick fix” to the problem is probably not very good. However, it might be possible to find a
circumvention that would allow a production application to continue running while you
investigate the memory leak. Consider the following scenarios:
򐂰 Did this apparent memory leak surface after an existing application was updated?
Potential circumvention:
– If possible, deploy the previous version of the application.
– Discuss the problem with the application developers that made the changes.
򐂰 Did the problem surface after i5/OS PTFs were installed?
Potential circumvention:
– Find out what PTFs were installed and contact IBM support to see if any problems
have been reported with those PTFs.
– Try to verify all properties or settings required by your application. If you have any sort
of configuration files stored in an i5/OS product directory, these might have been
overlaid by the installation of PTFs.

8.5.2 Memory leaks and the heap

The IBM Technology for JVM maintains two separate areas of memory. Together they
represent the total memory footprint. These are the Java heap and native heap. The Java
heap is the one that contains all the Java objects and is managed by the garbage collector. In
the context of Java, most references to “heap” are referring to Java heap. The minimum and
maximum size of the heap are configured using the -Xms and -Xmx settings.

The contents of the native heap are more stable than Java heap and are not subject to
garbage collection. The native heap is typically used when application JNI code does a
malloc(). Space for code that is optimized by the JIT compiler is also allocated from the native
heap. The size of the native heap is influenced by the -Xmx setting, because both heaps must
fit within a 32-bit address space, you can increase the size of the native heap by decreasing
the amount used by the Java heap, using the -Xmx setting.

Chapter 8. Analyzing JVM behavior 111

 A memory leak is much more likely to occur in the garbage-collected Java heap, therefore,
the remaining sections focus on the Java heap. For more information about native heap
usage on i5/OS we suggest you review Chapter 17 in the IBM Developer Kit and Runtime
Environment, Java 2 Technology Edition, Version 5.0 Diagnostics Guide on the Web at:
http://download.boulder.ibm.com/ibmdl/pub/software/dw/jdk/diagnosis/diag50.pdf

8.5.3 Collecting heap information

If none of the quick fix recommendations helped resolve your memory leak, then you have to
investigate what types of objects are accumulating in the Java heap. This is done by
generating heap dumps for analysis. The following instructions and sample commands
describe how the heap dumps are generated. In 8.5.4, “Using the Memory Dump Diagnostic
for Java (MDD4J) tool” on page 114 we describe a tool that is used to automate analysis of
heap dumps.

Triggered by an OutOfMemoryError
If you are familiar with the Classic JVM on i5/OS you probably are not accustomed to seeing
an OutOfMemoryError very often. The reason for this is that the default setting for maximum
heap size is essentially *NOMAX. The default maximum heap size for IBM Technology for
JVM is 2 GB. Because the default setting is defined and users of IBM Technology for JVM are
likely to tune the maximum heap size, the odds of seeing an OutOfMemoryError increases.
When the VM reaches an out of memory condition a heap dump is generated. As you can see
in Figure 8-24, the location of the dump file is logged to stderr.

JVMDUMP006I Processing Dump Event "uncaught", detail "java/lang/OutOfMemoryError" - Please

Wait.
JVMDUMP007I JVM Requesting Snap Dump using
'/home/pauls/java/Snap0001.20061006.141801.9909.trc'
JVMDUMP010I Snap Dump written to /home/pauls/java/Snap0001.20061006.141801.9909.trc
JVMDUMP007I JVM Requesting Heap Dump using
'/home/pauls/java/heapdump.20061006.141801.9909.phd'
JVMDUMP010I Heap Dump written to /home/pauls/java/heapdump.20061006.141801.9909.phd
JVMDUMP007I JVM Requesting Java Dump using
'/home/pauls/java/javacore.20061006.141801.9909.txt'
JVMDUMP010I Java Dump written to /home/pauls/java/javacore.20061006.141801.9909.txt
JVMDUMP013I Processed Dump Event "uncaught", detail "java/lang/OutOfMemoryError".
Exception in thread "main" java.lang.OutOfMemoryError
at java.util.HashMap.resize(HashMap.java:499)
at java.util.HashMap.addEntry(HashMap.java:800)
at java.util.HashMap.put(HashMap.java:441)
at java.util.HashSet.add(HashSet.java:209)
at MyLeakPgm.leak(MyLeakPgm.java:10)
at MyLeakPgm.main(MyLeakPgm.java:18)
Figure 8-24 A heap dump is generated by IBM Technology for JVM automatically on OutOfMemoryError

Capturing a heap dump manually

It is often very useful to take heap dumps at specific times or intervals so that you can see
changes in the Java heap over time. You can do this easily by sending a signal to the JVM,
but it does require that you start the JVM with the -Xdump:heap command-line option.

112 IBM Technology for Java Virtual Machine in IBM i5/OS

Figure 8-25 shows how this can be done when invoking Java from within a Qshell session.

QSH Command Entry

$
> java -Xdump:Heap SleepTimer 400
Figure 8-25 Setting the JVM to generate a heap dump

From a separate Qshell session you can use the ps utility to list all active processes running
under a given user profile. Figure 8-26 shows an example. There is a job
(204462/pauls/qp0zspwp in our example) that corresponds to the Java application. Note that
the process ID (PID) is 298.

QSH Command Entry

$
ps -u pauls
PID DEVICE TIME FUNCTION STATUS JOBID
299 - 00:00 pgm-ps run 204463/pauls/qp0zspwp
262 - 00:00 pgm-qzshsh evtw 204350/pauls/qzshsh
228 qpadev0006 00:00 cmd-qsh deqw 204242/pauls/qpadev0006
263 qpadev0008 00:00 cmd-qsh dspw 204351/pauls/qpadev0008
297 - 00:00 pgm-qzshsh evtw 204461/pauls/qzshsh
298 - 00:01 - thdw 204462/pauls/qp0zspwp
$
kill -QUIT 298
$
Figure 8-26 Find the PID and start the Javadump

Figure 8-26 also shows how to use the kill command to generate the Javadump. A
Javadump captures the current state of the JVM and can be automatically triggered under
certain error conditions. The dump includes basic information about garbage collection, lock
information, heap statistics, and a stack for each thread. In our example, (Figure 8-25) the
JVM was started with the -Xdump:heap option, so that a Heap dump is generated in addition to
the Javadump.

Important: Do not let the terms “kill” and “quit” confuse you. The VM continues running
after the kill command is issued. However, generating a heap dump (when JVM is
started with the -Xdump:heap command-line option) might take several minutes and it could
have a noticeable impact on your application. For this reason it is advisable that you only
generate heap dumps in a development environment. If the memory leak is only noticeable
when running an application in a production environment, then you have to plan for a time
that has the least impact to your business if an outage were to occur.

You might have noticed in Figure 8-26 that there is no confirmation message after the kill
command is issued.

Chapter 8. Analyzing JVM behavior 113

 As you can see in Figure 8-27, the output of your JVM is updated to include the location and
name of the dump files. This same information can also be found in the job log.

QSH Command Entry

java -Xdump:Heap SleepTimer 400

Going to sleep for 400 seconds...
JVMDUMP006I Processing Dump Event "user", detail "" - Please Wait.
JVMDUMP007I JVM Requesting Heap Dump using
'/home/pauls/heapdump.20061009.113847.298.phd'
JVMDUMP010I Heap Dump written to /home/pauls/heapdump.20061009.113847.298.phd
JVMDUMP007I JVM Requesting Java Dump using
'/home/pauls/javacore.20061009.113847.298.txt'
JVMDUMP010I Java Dump written to /home/pauls/javacore.20061009.113847.298.txt
JVMDUMP013I Processed Dump Event "user", detail "".
Figure 8-27 Showing the files generated when a heap dump is triggered

Deciding when to generate a heap dump

Because generating a heap dump can have a significant impact to the running application,
you must try to generate as few as possible. Here are a couple of tips that might help:
򐂰 Increase the maximum heap size of the JVM to allow the leak to grow bigger.
򐂰 Enable verbose GC output (see 5.2.3, “Verbose GC output” on page 44).
򐂰 Analyze verbose GC output to observe how fast the free heap space is dropping.
򐂰 Do not take any dumps when the application is in its “startup” phase.
򐂰 Take four to six memory dumps during the period that the free heap space is dropping.

8.5.4 Using the Memory Dump Diagnostic for Java (MDD4J) tool
Now that you have generated some heap dumps you have to analyze them to see if you can
get closer to finding a root cause of the memory leak. These dumps are much too big and
cryptic to try any kind of manual analysis, therefore, a tool is necessary to process and
summarize the data. There are a few tools that can do such analysis but we are going to
focus on one called Memory Dump Diagnostic for Java (MDD4J). The MDD4J tool combines
many of the best features from existing tools, such as Leakbot and HeapAnalyzer. IBM is
currently committed to making MDD4J the primary tool for performing memory leak analysis.

Note: The Memory Dump Diagnostic for Java tool has the following recommendations for
minimum hardware specifications:
򐂰 Approximately 5 GB of disk space
򐂰 1.5 GB of physical memory (RAM)
򐂰 2 GHz CPU (for Intel processors)

Installing MDD4J
MDD4J is available through the IBM Support Assistant tool. IBM Support Assistant (ISA) is
the mechanism for delivering and maintaining tools such as MDD4J. If you do not have ISA
installed or if you would like to read more about it, refer to Appendix B, “IBM Support
Assistant” on page 181.

114 IBM Technology for Java Virtual Machine in IBM i5/OS

Launching MDD4J
With the tool successfully installed you are ready to analyze some heap dumps.
1. Launch IBM Support Assistant.
2. Click Tools on the IBM Support Assistant Welcome Page.
3. Click WebSphere Application Server V6.1 under Products.
4. Click Memory Dump Diagnostic for Java (MDD4J) as shown in Figure 8-28.

Figure 8-28 Launching MDD4J tool

5. MDD4J must launch and look like Figure 8-29.

Figure 8-29 MDD4J start page

Chapter 8. Analyzing JVM behavior 115

 Analyzing data with MDD4J
MDD4J processes data and presents the results in the GUI. The following example helps
illustrate how the analysis is done and how you can interpret those results. To get started
Figure 8-30 shows a pictorial view of our Java program that leaks string objects.

Leak Root MyLeakPgm

Owner
Chain
Container HashSet

Leaking HashMap
Unit

HM$Entry HM$Entry HM$Entry

String String String

Contents Char[] Char[] Char[]

Figure 8-30 Showing the structure of the sample leak program

Defining the following terms help you understand Figure 8-30:

򐂰 Leak root: the class object which holds a static reference to the chain of objects leading to
the leak container (MyLeakPgm)
򐂰 Leak container: the object (HashMap) holding all the leaking objects
򐂰 Leak unit: the data type that is the root of the data structures which are added to the
enclosing Leak Container object (String Objects)
򐂰 Owner chain: is defined to be the chain of objects starting from a root

Notice in Figure 8-31, the MDD4J tool can accept two files as input. For a good comparative
analysis, primary and baseline dumps are collected during a single run of a memory leaking
application. The second file is optional and is used to do comparative analysis to the primary
file. The primary dump refers to the dump taken after the memory leak has progressed
considerably, consuming a large amount of the maximum configured heap size. The baseline
dump is captured early on, when the heap has not yet been consumed significantly due to the
memory leak. Because a comparative analysis typically yields the best results we show a
comparative analysis in our example.

116 IBM Technology for Java Virtual Machine in IBM i5/OS

See Figure 8-31.

Figure 8-31 MDD4J can accept two files as input for comparative analysis

We generated two heap dumps using the instructions in “Capturing a heap dump manually”
on page 112 while the sample leak program was running and specified them in the Baseline
and Primary dump file fields. When you click the Upload and Analyze button you must see
the screen shown in Figure 8-32.

Figure 8-32 The analysis status screen

Chapter 8. Analyzing JVM behavior 117

 Attention: The analysis could take a long time to finish. On a test PC (1.5 GHz, 1 GB of
RAM) this example took about 32 minutes to complete. The primary dump file was about
42 MB and the baseline dump file was 15 MB in size. At the time of this writing the MDD4J
development team was working on performance enhancements. All updates to MDD4J are
delivered via IBM Support Assistant. It is recommended that you regularly check for
updates and install any available updates.

If the analysis does not seem to be making any progress, you might want to look in the
following directory to see if there are any errors in the log files:
<IBM Support Assistant install>/workspace/logs

The default workspace directory location is shown in Figure 8-33. Some potential errors are:
򐂰 The dump file was truncated or corrupted in some way
򐂰 There are too many objects in the dump or the size of the dump exceeded what MDD4J
can process. Currently MDD4J can process dumps containing up to 30 million objects.

Figure 8-33 Location of the MDD4J log and results files

When the analysis does finish a summary screen is displayed as shown in Figure 8-34.

Figure 8-34 Summary of a comparative analysis

118 IBM Technology for Java Virtual Machine in IBM i5/OS

There are many different views of data and you probably find a few that you prefer as you
work with the tool. Exploring the Next Steps section on the Analysis summary tab, shown in
Figure 8-35, is a good way to see the various summaries and views available.

Figure 8-35 Following the Next Steps to view the analysis results

The Suspects view, shown in Figure 8-36, is a great place to start looking because, as the
name implies, it indicates likely causes of a memory leak.

Figure 8-36 Showing suspected leak information

Chapter 8. Analyzing JVM behavior 119

 A graphical view is also available, as shown in Figure 8-37, by clicking the Explore Context
and Contents tab.

Figure 8-37 A graphical view of ownership context

You can also see details about selected objects in the tree by switching to the Browse tab.

In Figure 8-38 the details of the class MyLeakPgm are displayed. You can see that the total
reach size is about 98 MB. This is a very big clue because you already know that the total
heap size was about 105 MB when the second heap dump was generated, as shown in
Figure 8-34 on page 118. This means that nearly all the heap space is “rooted” in the
MyLeakPgm class.

This is a very obvious example of a memory leak but it is this type of information that would
allow a Java developer to locate and fix the faulty code.

Note: Total reach size is defined as “The size of all objects which are reachable from the
given object, in a single pass depth-first search of all the objects in the memory dump
starting from all root objects where any object is visited at most once”.

120 IBM Technology for Java Virtual Machine in IBM i5/OS

 See Figure 8-38.

Figure 8-38 Showing details of selected objects in the tree

If you would like to view the heap contents directly you can bring up the object table in a
sortable view. For example, you can view the object table by performing the following steps:
1. Start at the Analysis summary page, shown in Figure 8-34 on page 118, and click To see
all the objects and object types in the heap in a sortable and tabulated format: Click
here.
2. When the object table loads click the Types Table tab.
3. Click the Growth in instances since Baseline dump column.
The object table is now sorted and showing the object types that have grown the most
since the baseline dump was taken.

Chapter 8. Analyzing JVM behavior 121

 See Figure 8-39.

Figure 8-39 Showing the object table

The Future of MDD4J

As mentioned in 8.5.4, “Using the Memory Dump Diagnostic for Java (MDD4J) tool” on
page 114, IBM is currently committed to supporting this tool as the primary customer tool for
analyzing memory leaks. Enhancements and fixes for MDD4J are released on a regular
basis, therefore, be sure that you are using the “Updater” feature of IBM Support Assistant to
download and apply the fixes to your workstations. If you want to report a problem with
MDD4J or any other IBM Support Assistant tool, use the IBM Support Assistant feedback
option. For additional information about MDD4J visit the following Web site:
http://www-128.ibm.com/developerworks/websphere/library/techarticles/0606_poddar/0
606_poddar.html

Analyzing memory leaks with WebSphere Development Studio Client for

iSeries
WebSphere Development Studio Client for iSeries Advanced Edition has an option for
analyzing memory leaks. It allows you to take snapshots of the memory heap automatically or
manually as your application runs. You can then analyze the number of objects and memory
usage growth during the interval. It includes several useful views, such as:
򐂰 Leak candidate view for identifying the objects most likely responsible for memory leaks
򐂰 Object reference graph, object references, and object details views provide information
about objects consuming memory
򐂰 Statistical views display profiling statistics for packages, classes, and methods

122 IBM Technology for Java Virtual Machine in IBM i5/OS

In this section you can see how to identify a memory leak. We use the Identify Java memory
leaks sample that comes with the tool.

Restriction: At the time of this writing, you can only use the Classic JVM for remote
debugging in WebSphere Development Studio Client for iSeries version 6.0.1. However
the approach presented in this section is applicable to IBM Technology for JVM when this
is supported for remote debugging.

Perform the following steps:

1. Start WebSphere Development Studio Client for iSeries Advanced Edition on your
workstation.
2. Open the Java perspective.
3. Locate the Java application project and source files for your application.
4. Right mouse click the class (which must have a main() method) and select Profile → Java
Application.
5. At the properties window select the Profiling tab as shown in Figure 8-40).
6. Select the Memory Leak Analysis - Manual heap dumps option and click OK.

Figure 8-40 Memory leak analysis selection

7. If you do not have the profiling and logging perspective open, then you can see a confirm
perspective switch? window open. If so, click Yes.
8. You must be in the profiling and logging perspective and see your application status as
monitoring. If the console does not open go to the toolbar and select Window → Show
View → Console. This displays any output data.
You have to manually capture at least two heap dumps to do memory leak analysis.

Chapter 8. Analyzing JVM behavior 123

 9. After the application has “warmed up”, you have to capture a heap dump. Click the
Capture Heap Dump button as shown in Figure 8-41.

Figure 8-41 Manually capture a JVM heap dump

10.You must see a heap dump listed under your running application. Let the application
continue running.
11.After the application has run a few more times, click the Capture Heap Dump button
again.
12.You must see two heap dumps, as shown in Figure 8-42. Click the Terminate button to
stop the application and the profiling.

Figure 8-42 heap dumps available for analysis

124 IBM Technology for Java Virtual Machine in IBM i5/OS

13.You must see the leak candidates view. Click the Analyze for Leaks button. See
Figure 8-43.

Note: The memory leak analysis tools use technology from MDD4J. You can import
and analyze heap dumps from standard heap dumps (using the IBM_HEAPDUMP
environment variable, an HPROF heap dump, an OS/390® heap dump, or a Hyades
optimized heap dump.

Figure 8-43 Analyze the heap dumps for memory leaks

14.At the “select leak analysis options” window you must see your two heap dumps selected.
You must also see the threshold set to 20 (the default). Setting the number low attempts to
find many leaks. Setting the number high attempts to find fewer. In this example the
default is used. Click OK after specifying the appropriate heap dumps and threshold.
It might take a few minutes to analyze the heap dumps. You must see the progress
statistics in the lower right part of the window, as shown in Figure 8-44.

Figure 8-44 Analyzing memory leak candidates

If one or more memory leaks have been detected, they are listed in the leak candidates
view. The higher the likelihood value, the higher the probability that a memory leak has
been detected. You must also see the parent application, the object type causing the leak
and the amount of memory.

Chapter 8. Analyzing JVM behavior 125

 In this example you see from Figure 8-45 that a vector is holding on to references for
String objects.

Figure 8-45 Memory leak candidate identified

15.Double click the line with the leak candidate data entries. You must see an object
reference view open, with the “path” to the offending container and object. If you hover the
mouse over an object or the link, you see statistics on the number of objects and memory
leaked, as shown in Figure 8-46.

Figure 8-46 Object reference view of the potential memory leak

In this example the memory leak is in the SecondaryQueue class, from a vector object that
is retaining references to String objects.
16.The next step is to go back to the Java perspective and open up the source file for the
offending class, as shown in Example 8-2. This is the offending code. myQ is a vector
object and retains references to all objects that it contains.

Example 8-2 Source code

private Vector myQ;
private int currentPos;
public SecondaryQueue(){
myQ = new Vector();
currentPos=0;
}

126 IBM Technology for Java Virtual Machine in IBM i5/OS

 public void add(Object obj){
myQ.add(obj);

public Object getNext(){

// First In First Out.
// Get an object exactly once,
// currentPos keeps track of last item removed.
if(myQ.size()>currentPos){
currentPos++;
return myQ.get(currentPos-1);
}
return null;
}

17.Look at the add(Object obj) method. Each invocation of this method adds an object to
the myQ vector. These could be String or other objects. As objects are added to the
vector, the memory usage increases accordingly.
18.Look at the getNext() method. This returns the object at currentPos, but the object still
stays in the myQ vector. This is a form of a memory leak. In order to properly remove an
object from Vector, one must use the remove method:
public E remove(int index)
This method removes an object from the Vector and returns it to the caller.
The next code snippet shown in Example 8-3 illustrates how to use this to fix the memory
leak. In this case the object is removed from the Vector after the getNext() method. Now
the myQ vector object cannot grow indefinitely like it could in the previous example.

Example 8-3 Correct implementation of the getNext method

public Object getNext(){
if(myQ.size()>0){
return myQ.remove(0);
}

return null;
}

8.6 Performance issues

The focus of this book is JVM specific issues. Other performance issues related to an
application or middleware product such as WebSphere Application Server are not discussed
here, but we provide links to more detailed resources related to these performance issues.

8.6.1 Quick fix: general guidelines

Creating a JVM process to run Java applications on System i platform requires memory and
process resources. For short-running applications it takes longer to create the JVM process
than the actual application run time. Middleware products such as WebSphere Application
Server create a JVM process and reuse it for all runtime operations.

Chapter 8. Analyzing JVM behavior 127

 A common approach to minimizing the impact of JVM creation for System i platform is to write
the Java application such that it listens for requests on an i5/OS data queue and places any
output on another data queue. In other words, the requesting application sends a message to
a data queue with the appropriate input parameters, rather than calling the application
directly.

Generally speaking, performance problems can be caused by the client, network, or server.
Common client problems include old hardware, inadequate memory, missing fixes, and
poorly written code. Common network problems include inadequate bandwidth, latency,
excessive broadcasting or file transfer activity, and poor design. Your first step in dealing with
perceived performance issues must be to investigate whether the client or network is an
issue, and then look at potential issues at the server level.

8.6.2 Server guidelines

These are the major categories of server based contributors to overall Java application
performance on System i platform:
򐂰 Hardware
򐂰 Operating system and system values
򐂰 Database or other i5/OS resources
򐂰 Coding techniques
򐂰 Java virtual machine

The Java virtual machine guidelines are covered next in 8.6.3, “Java virtual machine” on
page 132.

Hardware
General guidelines for hardware include having adequate processor, memory, disk, and
communications resources. System i IBM POWER™ 5 and IBM POWER 5+ based servers
provide faster response time and more throughput than previous generation IBM POWER 4
based servers. You can use i5/OS commands such as WRKACTJOB, WRKSYSSTS, and
WRKDSKSTS to perform real time sampling to understand processor, memory, and disk
usage on your System i platform. You can also use i5/OS Performance Collection Services to
perform long term monitoring.

You can use the IBM Systems Workload Estimator (WLE) to size a new system, a proposed
upgrade to an existing system, or a consolidation of several systems. The WLE tool currently
does not have the capability to size generic Java workloads. However, it can size WebSphere
Application Server-based workloads and provide basic guidelines on processor, memory, and
disk requirements for a given workload. WLE is available at no charge on the Web at:
http://www-912.ibm.com/supporthome.nsf/document/16533356

Java applications tend to consume more memory and processor resources than compiled
applications written in RPG or COBOL. Java applications also tend to use more high level
language instructions than traditional System i applications and benefit from newer System i
models. The IBM Accelerator for System i5 is available on selected System i5 520 models
and can also significantly improve Java application performance. The accelerator is a
software feature code that enables the entire processor’s capacity to be available for batch
processing workloads. For example, feature number 7354, available for the System i 7143
520 Express Configuration, increases the commercial processing workload (CPW) rating for
System i platform from 1200 to 3800. Another key recommendation is for Java based
applications to be run on systems with L3 cache. Refer to the IIBM System i5 Handbook IBM
i5/OS Version 5 Release 4 January 2006, SG24-7486.

128 IBM Technology for Java Virtual Machine in IBM i5/OS

For additional recommendations visit the System i performance management Web site. It
includes links to generic System i performance resources such as benchmark results and
specific articles on Java and WebSphere Application Server for i5/OS. It also has links to the
i5/OS performance capabilities and references guides, which provides extensive tips and
techniques for optimizing overall System i performance. These resources are available at:
http://www-03.ibm.com/servers/eserver/iseries/perfmgmt/resource.html

Operating system and system values

As mentioned in the previous section, Java applications tend to consume more processor and
memory resources than traditional System i applications. Some of this is an inevitable
trade-off because of the code portability Java applications provide. More importantly, Java
applications can provide function and integration that is difficult or impossible with traditional
System i applications. This must more than compensate for the additional system resource
requirements.

Running Java based applications on System i platform might require you to change your
i5/OS work management and tuning strategies. Java is inherently a multithreaded
environment and the JVM spawns threads based on how the application was written and
other runtime events. Java based middleware such as WebSphere Application Server also
makes extensive use of threads for its internal operations, and also accessing System i
applications and data resources. An execution thread on System i platform is represented as
an activity level. Activity levels can be specified at multiple levels (memory pool, subsystem,
and so on).

In general, the activity level must be high enough to avoid wait → ineligible and active →
ineligible transitions, but not so high as to cause system instability. You can use the i5/OS
WRKSYSSTS command to view the current activity level in your memory pools. Refer to the
Max Active column shown in Figure 8-47.

Work with System Status RCHAS60

10/09/06 11:23:38
% CPU used . . . . . . . : .6 Auxiliary storage:
% DB capability . . . . : .0 System ASP . . . . . . : 492.3 G
Elapsed time . . . . . . : 00:00:00 % system ASP used . . : 23.4057
Jobs in system . . . . . : 979 Total . . . . . . . . : 597.8 G
% perm addresses . . . . : .010 Current unprotect used : 8559 M
% temp addresses . . . . : .028 Maximum unprotect . . : 9162 M

Type changes (if allowed), press Enter.

System Pool Reserved Max Active-> Wait-> Active->

Pool Size (M) Size (M) Active Wait Inel Inel
1 1394.55 691.16 +++++ .0 .0 .0
2 25107.60 4.20 590 1132 .0 .0
3 3193.59 .00 299 125.7 .0 .0
4 .25 .00 5 .0 .0 .0

Bottom
Figure 8-47 WRKSYSSTS screen shows the size and activity levels for each memory pool

Chapter 8. Analyzing JVM behavior 129

 By default, Java applications run in the System i *BASE memory pool. Therefore, you must
ensure that the activity level is set appropriately, and that there is adequate memory in the
pool. This can be influenced (positively or negatively) by the setting of the Performance
adjustment (QPFRADJ) system value. If this is set to a value other than 0, i5/OS attempts to
periodically move memory and activity levels between memory pools. This might or might not
be advantageous for your particular environment. A common recommendation is to
appropriately tune the memory pool size and activity level for all workloads on your systems,
then set the QPFRADJ system value to 0 (no adjustment). You can then monitor your system
resource usage, throughput and other performance results and make changes as
appropriate. However, this might not always be possible in production environments.

A common technique is to isolate Java (especially WebSphere Application Server) based

workloads into their own memory pool. This can be particularly advantageous in
environments where multiple workloads are competing for the same processor, memory, and
activity level resources. i5/OS allows you to create a shared or private memory pool and
define the pool size and activity level to ensure your Java applications always have an
appropriate level of resources. i5/OS also allows you to create your own subsystems and
define routing entries that specify the memory pools to be used for specific tasks. This is also
popular on System i platform, especially with WebSphere Application Server based
workloads. It is likely that you have to evaluate the feasibility of doing this in your environment
on a case by case basis, especially if you have limited memory resources on your System i
platform.

There are many other i5/OS tuning tips and techniques you can use, such as ensuring the
prestart jobs for database requests (job QSQSRVR in the QSYSWRK subsystem and job
QZDASOINIT in the QUSRWRK subsystem) are set high enough. For a comprehensive list of
recommendations on i5/OS tuning for Java workloads, refer to chapter 5 of the IBM Redbook
titled Maximum Performance with WebSphere Application Server V5.1 on iSeries,
SG24-6383.

Database and other i5/OS resources

Another key element in obtaining the best possible performance from Java applications
running on i5/OS, is to ensure you have optimized access to resources such as DB2 for
i5/OS. This includes optimizing:
򐂰 Connection to the database
򐂰 Java database connection (JDBC) and SQL code
򐂰 Database design

Generally speaking, Java based applications use a Java Database Connectivity (JDBC)
driver to connect to relational database resources such as DB2 for i5/OS. There are two
JDBC drivers available for DB2 for i5/OS:
򐂰 The native driver
򐂰 The IBM Toolbox for Java driver

Generally speaking, the toolbox driver is more flexible but the native driver performs better.
Also, an XA enabled JDBC driver (for transactional integrity such as two phase commit)
imposes additional overhead compared with a non-XA enabled JDBC driver.

130 IBM Technology for Java Virtual Machine in IBM i5/OS

Table 8-1 provides a comparative analysis of two JDBC drivers.

Table 8-1 Comparison between the System i native and toolbox drivers
Attribute Native JDBC driver IBM Toolbox for Java JDBC
driver

Implementation Database and Java application Database and Java application can
must be on the same machine be on the same or different
or partition. partitions.

Packaging Packaged with product Packaged with product 5722-JC1 or

5722-JV1. available at:
http://jt400.sourceforge.net/

Application usage Can be used with JDBC driver Can be used with JDBC driver
manager or data source manager or data source
connections. connections.

Transaction integrity XA and non-XA enabled XA and non-XA enabled versions

versions available. available.

i5/OS job structure Uses QSQSRVR jobs in Uses QZDASONIT jobs in

QSYSWRK subsystem. QUSRWRK subsystem.

Interface to DB2 for i5/OS Uses direct call level interface Uses database host server job and
to DB2 for i5/OS. TCP/IP sockets interface to DB2 for
i5/OS.

Coding techniques
Another key to optimizing performance is coding techniques related to DB2 access. For
overall Java coding recommendations refer to chapter 9 in the IBM Redbook titled Maximum
Performance with WebSphere Application Server V5.1 on iSeries, SG24-6383.

Using a data source connection within your application (rather than the JDBC driver
manager) allows you to take advantage of database connection pooling. Connection pooling
provides a set of reusable i5/OS database access jobs that improves scalability. Connection
pools also facilitates the reuse of prepared statements and open data paths, which can also
improve scalability. Both the native and toolbox JDBC drivers support connection pooling.

Stored procedures are another option for improving database access performance. Stored
procedures represent callable programs on the System i platform, which improve
performance in several ways. Stored procedures help reduce the number of calls between
the application and the DB2 for i5/OS runtime environment. Stored procedures also minimize
the overhead associated with dynamic Structured Query Language (SQL) such as parsing
and validating SQL statements, and creating an access plan. The toolbox JDBC driver
supports SQL packages, which can also improve overall scalability of your applications.
Efficient SQL coding techniques, such as only accessing the data you require, are also
essential. For examples, refer to section 9.2.4 of the IBM Redbook titled Maximum
Performance with WebSphere Application Server V5.1 on iSeries, SG24-6383.

Finally, ensure you have good database design concepts in place, such as indexes. DB2 for
i5/OS supports binary radix indexes, and also encoded vector instances. Depending on your
specific database environment and SQL statements used in your applications, one might be
much preferable over the other. You can use the Visual Explain tools (part of the iSeries
Navigator product) to create and analyze database monitor jobs on your System i platform.

Chapter 8. Analyzing JVM behavior 131

 iSeries Navigator tools also include index advisors, which suggest the most appropriate types
of indexes over your table columns. You can also use the Visual Explain tools to model and
optimize your SQL statements. Additionally, ensure you are using the latest versions of the
DB2 for i5/OS group PTF package (SF99504 on V5R4M0).

For more detailed information about DB2 for i5/OS performance, the following publications
are helpful:
򐂰 Preparing for and Tuning the SQL Query Engine on DB2 for i5/OS, SG24-6598
򐂰 Stored Procedures, Triggers, and User-Defined Functions on DB2 Universal Database for
iSeries, SG24-6503
򐂰 SQL Performance Diagnosis on IBM DB2 Universal Database for iSeries, SG24-6654

8.6.3 Java virtual machine

This section considers how JVM features such as JIT compilation and garbage collection can
influence application performance. Garbage collection is an area which can influence
application performance to some extent. Chapter 5, “Tuning the garbage collector” on
page 35 discussed how to adjust garbage collection settings appropriate to your environment.
This includes steps such as setting the initial heap size and maximum heap size to
appropriate values. This section takes an example of an application that is performing poorly.
We show how to investigate if inappropriate garbage collection settings are having a negative
impact on application performance and how to confirm that the problem is resolved.

It is worth repeating that for the majority of applications the default GC policy gives
satisfactory performance. Adjusting the GC policy must be done only when your application
performance does not meet your goals. Keep in mind though that you have to perform tuning
activities at the application level first, for example, based on application profiling.

The performance gain to be achieved by using an alternative garbage collection policy varies
by application, but can be quite small. In some cases a particular GC policy might in fact have
a negative impact on application performance.

General guidelines
The IBM Technology for JVM has many built-in performance features, such as an advanced
JIT compiler that optimizes code. For example, you do not have to “tune” the JIT threshold
parameter. In Classic JVM, you can set the os400.jit.mmi.threshold=xyz custom property to a
value other than the default of 2000, however, this applies to the entire JVM and is not widely
used in practice. Refer to chapter 5 of IBM Developer Kit and Runtime Environment, Java 2
Technology Edition, Version 5.0 Diagnostics Guide, for details on the IBM Technology for
JVM JIT optimization.

An important aspect of JVM performance is to ensure your applications do not have a

memory leak or that they do not indiscriminately create objects. A memory leak occurs when
objects improperly hold on to references, such as can happen in recursive calls. It can cause
memory usage to continue growing over time even though the workload has reached a
steady state condition. Refer to 8.5, “Investigating a suspected memory leak” on page 111 for
more details.

Verbose garbage collection typically imposes modest overhead and provides useful JVM
runtime information. However, verbose class loading and verbose JNI impose overhead and
are most useful for debugging, rather than runtime performance analysis. You can also adjust
the size of your shared class cache, as described in 6.2.2, “Deploying shared classes” on
page 54.

132 IBM Technology for Java Virtual Machine in IBM i5/OS

Some other general recommendations are:
򐂰 Do ensure you are using the latest Java group PTF package (SF99291 for i5/OS
V5R4M0)
򐂰 Do not disable the JIT compiler unless performing troubleshooting
򐂰 Do not enable JVM profiling or other tracing unless it is absolutely necessary
򐂰 Do not disable class garbage collection (JVM argument -Xnoclassgc)

Important: Be very careful when you specify JVM command line arguments. Mistakes
as simple as case sensitivity in the parameters can prevent the JVM from starting.

Troubleshooting garbage collection

5.2.3, “Verbose GC output” on page 44 explained how to manually analyze verbose GC
output to support heap analysis and tuning activities. The IBM Support Assistant provides a
verbose GC data visualizer tool which you can use for such a task; the Extensible Verbose
GC Toolkit (EVTK).

Overview of EVTK
EVTK parses and plots verbose GC logs and -Xtgc output. It provides graphical display of a
wide range of verbose GC data values and at the time of writing is compatible with optthruput,
optavgpause, and gencon GC policies.

You can use EVTK to support GC tuning and also general GC analysis and troubleshooting.
We present a scenario where EVTK is used to analyze verbose GC output from a
WebSphere Application Server environment which is not performing as well as expected.
After identifying the possible cause of the performance bottleneck from the EVTK analysis,
we make changes and use EVTK to confirm that the problem has been resolved.

Note: At the time of writing this book IBM was planning to make EVTK available in IBM
Support Assistant.

The scenario presented here involved a WebSphere Application Server V6.1.0.1 Base server
with the Trade 6 performance sample application installed.

Trade 6 application
IBM Trade Performance Benchmark Sample for WebSphere Application Server (otherwise
known as Trade 6) is the WebSphere end-to-end benchmark and performance sample
application. The new Trade benchmark has been re-designed and developed to cover
WebSphere's significantly expanding programming model. This provides a real world
workload driving WebSphere's implementation of Java 2 Enterprise Edition (J2EE) 1.4 and
Web Services, including key WebSphere performance components and features.

Test scenario
To stress the application server we generated a simulated workload for the Trade 6
application using Rational® Performance Tester V6.0.0.1. We configured verbose GC output
prior to starting WebSphere Application Server by specifying the following filename in the
Generic JVM arguments field:
verbosegclog:filename

Refer to “Configuring the WebSphere Application Server JVM” on page 173 for instructions
on setting JVM arguments.

Chapter 8. Analyzing JVM behavior 133

 We used the default optthruput GC policy and initially ran the scenario with artificially small
heap size settings so that the garbage collector would have to work very hard to prevent the
heap from running out of space completely:
-Xms35m -Xmx60m

These parameters were also set in the Java Virtual Machine settings of our test server.

The resource constraints in this scenario are clearly highly exaggerated for the purposes of
demonstrating the usage of EVTK for analyzing verbose GC data. Correspondingly, the
results are not typical, but show that for a highly-loaded and constrained application, GC
pauses can have a measurable impact on application throughput and response time, as you
might expect. EVTK allows us to determine that on this occasion the cause of poor
performance is related to increased garbage collection due to insufficient resources.

The server was placed under load by Rational Performance Tester for ten minutes and the
verbose GC log collected for analysis. Because garbage collection cycles occurred prior to
test start, they were not due to the test workload, data for these collections was manually
removed from the verbose GC output before analysis with EVTK. This was necessary so that
the graphs produced by EVTK and Rational Performance Tester covered the same period
and could therefore be compared like-for-like.

It is possible to change the period of verbose GC data displayed in EVTK, but you would have
to adjust both the start and end times displayed through EVTK to correlate to the start of the
test (as opposed to the startup of the JVM) for the same reason. We felt it was more accurate
to edit the verbose GC log file manually because we were able to refer to the time stamp in
each GC cycle entry and tell for certain whether it occurred during the test period.

Viewing verbose GC data in EVTK

At the time of this writing, EVTK was not yet integrated into IBM Support Assistant and was
supplied as a JAR file.

To view the verbose GC data in EVTK, we used the following procedure:

1. Copy the verbose GC output log file from i5/OS to a PC with the IBM Support Assistant
and EVTK installed.
2. Start EVTK by running the vtk.bat file.
3. Select File → Open File.
4. Locate the verbose GC log file to be analyzed then click Open.
EVTK processes the verbose GC log and displays a default chart from the data, as shown
in Figure 8-48.

134 IBM Technology for Java Virtual Machine in IBM i5/OS

 See Figure 8-48.

Figure 8-48 The default view displayed in EVTK after you open a verbose GC log file for analysis

The default graph did not meet our requirements. The data you want to visualize changes
during the troubleshooting process, but fortunately displaying new graphs in EVTK is
easy. You simply add or remove the data you want to plot through the VGC Data menu.
5. For instance, we wanted to look at how frequent and how long GC pauses were during the
test period. To do this starting from the default graph shown in Figure 8-48:
a. Select VGC Data then uncheck Tenured heap size from the menu, as shown in
Figure 8-49.
Notice EVTK dynamically updates the graph to reflect the selections in the VGC Data
menu.
b. Select VGC Data then uncheck Free tenured heap (after collection).
This gives a blank chart onto which you can plot the required data in the next step.
c. Select VGC Data then check Pause times (mark-sweep-compact) collections to
display GC pause data on the graph.
This results in the chart shown in Figure 8-49.

Chapter 8. Analyzing JVM behavior 135

 See Figure 8-49.

Figure 8-49 Deselect the Tenured heap size data to give a better scale for the free heap space data

Test results
The effect of insufficient space in the heap on application throughput, application response
times, and GC pause times is clear from Figure 8-50, Figure 8-51, and Figure 8-52.

Figure 8-50 shows that average response times increase as the workload on the WebSphere
Application Server increases.

Figure 8-50 Response times increase as the workload increases

136 IBM Technology for Java Virtual Machine in IBM i5/OS

 Note: The period of apparent response time decrease from approximately 400 seconds
after test start onward, is misleading. The server at this time was occupied by the GC
workload that tried to free up heap for new allocations. The response time in fact increased
but this was not detected by Rational Performance Tester.

Figure 8-51 shows that the throughput of the Trade 6 application drops as the workload on the
WebSphere Application Server increases, reaching a point of zero throughput when the
garbage collector is running very frequently to ensure the heap does not completely run out of
free space.

Figure 8-51 Throughput drops as response time (and GC pause times) increases

In Figure 8-52 you can see that as the workload increases, garbage collections occur more
frequently, and take longer. This is because of the very low maximum heap size. As the
workload increases, the garbage collector has to frequently compact the heap to prevent the
heap running out of free space completely:

Figure 8-52 Garbage collection pauses increase in length and frequency as workload increases

Chapter 8. Analyzing JVM behavior 137

 Further analyzing verbose GC data with EVTK
EVTK provides several ways of displaying verbose GC data. For example, data can be
exported in CSV format or displayed as a graph. You can chart a wide variety of GC data,
featuring just one value plotted against elapsed time (such as free heap space) or combining
multiple values on one chart. Typically this feature would be used to easily see the relation
between multiple data values, such as free heap space and GC pause times. It is also
possible to overlay data from multiple verbose GC logs on the same chart for comparative
analysis between JVMs run with different GC settings.

In our case, we initially created a chart of the GC pause times in EVTK from the verbose GC
data, using the steps in “Viewing verbose GC data in EVTK” on page 134. This clearly
showed that GC pauses were occurring more frequently and getting longer as the workload
increased. This lead us to suspect that the heap is running out of space under load.

You can use EVTK to easily verify your suspicion that the heap is running out of space.
Re-plot the chart in EVTK to show the free heap space remaining against time. This gives the
chart shown in Figure 8-53.

Figure 8-53 Plotting free heap size clearly shows that the heap is running out of space as workload increases

You can see that at the half-way point in the test, the available free heap space suddenly and
sharply decreases to very low levels and the JVM eventually runs out of heap space.

You must, therefore, adjust the heap settings as follows:

-Xms50m -Xmx256m

Restart WebSphere Application Server so that the new heap size settings take effect.

138 IBM Technology for Java Virtual Machine in IBM i5/OS

 Results with new heap size settings
Running the scenario again, the positive effect on application response time and throughput
can be clearly seen in Figure 8-54 and Figure 8-55.

Figure 8-54 Average response time decreases after adjusting heap settings to more appropriate values

The application throughput also increases with the new heap settings, as shown in
Figure 8-55.

Figure 8-55 Application throughput increases after adjusting heap settings to more appropriate values

Chapter 8. Analyzing JVM behavior 139

 Confirming improved GC performance with EVTK
We collected verbose GC output from the second run of the application to compare to the
previous analysis and confirm that GC is operating more efficiently.

EVTK allows you to plot data from multiple verbose GC logs simultaneously on the same
chart. This is very useful for comparison purposes. In our scenario we overlaid the free heap
data from the application run with new heap settings, over the data from the problematic run:
1. Copy the new verbose GC log file from i5/OS to the PC running EVTK.
2. Open the original verbose GC log file if it is not currently open and plot the “Free tenured
heap (after collection)” data only:
a. Select File → Open File.
b. Select the original verbose GC log file and click Open. The default graph of the original
data displays in EVTK showing Tenured heap size and Free tenured heap.
c. Select VGC Data, then uncheck Tenured heap size.
3. Add the data from the latest test with new heap settings, to the graph:
a. Select File → Add File.
b. Select the new verbose GC log file and click Open.

The new data is then displayed as the blue plot line in the chart shown in Figure 8-56.

2nd run with

new heap
settings

1st run with

constrained
heap settings

Figure 8-56 More free heap space is available after adjusting heap settings to more appropriate values

You can clearly see from the new data plot in Figure 8-56 that the heap no longer rapidly runs
out of space as the application workload increases.

140 IBM Technology for Java Virtual Machine in IBM i5/OS

8.6.4 Application guidelines
Application design and coding techniques can have a major impact on performance. This
section covers:
򐂰 Application coding tips and recommendations
򐂰 Application profiling and analysis: workstation
򐂰 Application profiling and analysis: server

Application coding tips and recommendations

One of the most common sources of application performance problems is excessive object
creation. String concatenation is a prime example. Strings are easy to use, but are immutable
(meaning every change involves another object creation). Each string object consumes
working memory and increases the CPU usage during the garbage collection cycle.

If you have a lot of string manipulations in your program, for example, concatenating three or
more strings, use StringBuffer as the target object for concatenated string. This is especially
important if you do string concatenation in a loop.

For example, instead of using String to hold the result:

String s1 = “Use ”;
....
s1 = s1 + “StringBuffer ”;
.....
s1 = s1 + “for “;
...
s1 = s1 + “concatenation!”;
System.out.println(s1);

Use the StringBuffer object:

StringBuffer sb = new StringBuffer(“Use “);
...
sb.append(“StringBuffer ”);
...
sb.append(“for “);
...
sb.append(“concatination!”);
System.out.println(sb.toString());

As the result, in the first code example 8 new objects have been created, while in the second
example only 5. If you have to run this code segment as part of the loop you would create 3 x
<number of iterations> less objects with the second approach.

Another source of application performance problems is using “expensive” tasks such as

synchronized methods and reflection. They can be very useful, just bear in mind that they
tend to be fairly resource intensive. Another typical error is in overusing exception blocks. Use
exception blocks for actual “exceptions”, not for controlling the application flow.

Application profiling and analysis: workstation

It is recommended thoroughly testing and profiling your application prior to putting it into
production under IBM Technology for JVM for i5/OS. However, it might be the case that
performance issues might not surface until the application is in production and placed under
heavy load. In this case it might be necessary to perform profiling activities to establish where
any bottlenecks might be occurring in the application.

Chapter 8. Analyzing JVM behavior 141

 IBM JVMs prior to JDK version 5.0 implemented the Java Virtual Machine Profiling Interface
(JVMPI). JVMPI provides an interface between the JVM and a profiling agent. IBM provides a
Remote Agent Controller (RAC) application as the profiling agent for its JVMs. The RAC is
available on multiple platforms, including i5/OS. When the JVM is started in profile mode, it
links with the RAC and notifies it of events such as method calls and memory allocation. The
RAC issues control requests through the JVMPI, such as what events to monitor.

The JVMPI has been deprecated in JDK version 5.0 in favor of the Java Virtual Machine
Tools Interface (JVMTI). JVMTI is the strategic interface for application monitoring and
performance analysis.

IBM JVMs use a remote agent controller server daemon to implement JVMPI and JVMTI.
These are available on a variety of platforms (including System i platform) and are currently
packaged with application development tools such as WebSphere Development Studio Client
for iSeries. There are two key client application interfaces to JVMPI and JVMTI:
򐂰 Tivoli Performance Viewer, which is integrated within the WebSphere Application Server
administration console
򐂰 Rational Software Development Platform profiling and logging perspective

Tivoli Performance Viewer provides basic information about response time for servlets,
JavaServer™ Pages™ (JSP™), and Enterprise JavaBeans™ (EJB™) methods, and also
detailed information about JVM statistics such as garbage collection. Refer to Appendix A,
“Running WebSphere Application Server with IBM Technology for JVM” on page 167 for more
details.

WebSphere Development Studio Client for iSeries Advanced Edition provides detailed
information about method execution time, memory usage, thread analysis and code usage in
J2EE and Java 2 Standard Edition (J2SE) applications. The next few sections illustrate how
to identify potential performance and functional problems in Java applications using
WebSphere Development Studio Client for iSeries.

Attention: At the time of this writing, WebSphere Development Studio Client for iSeries
was using JVMPI interface. However, the concepts must still apply to JVMTI when the
appropriate support is available in the tool.

Analyzing response time and memory usage

This example shows how you can use WebSphere Development Studio Client for iSeries
Advanced Edition to analyze an application’s response time components and determine
memory and object usage.

Restriction: At the time of this writing, only the Classic JVM could be used for remote
debugging in WebSphere Development Studio Client for iSeries version 6.0.1. However,
the approach presented in this section is applicable to IBM Technology for JVM when this
is supported for remote debugging.

Perform the following steps:

1. Start WebSphere Development Studio Client for iSeries Advanced Edition on your
workstation.
2. Open the Java perspective.
3. Locate the Java application project and source files for your application.

142 IBM Technology for Java Virtual Machine in IBM i5/OS

4. Right mouse click the class (which must have a main() method) and select Profile → Java
Application.
5. At the properties window select the Profiling tab. See Figure 8-57.

Figure 8-57 Application profiling selections

6. Select the Basic Memory Analysis option and click Edit.

7. At the Edit Profiling Set window enter a descriptive profiling set name and description.
Click Next.
8. Expand the Time Analysis section and select the Execution Time Analysis option.
9. Select the Collect method CPU time information option.

Chapter 8. Analyzing JVM behavior 143

 See Figure 8-58.

Figure 8-58 Monitor basic memory usage and method response times

10.Click Next and you must see the default filter set selected.
11.In the Contents of selected filter set section click the Add button.
12.Specify the package you want to monitor. You can use the wildcard character * to select
all classes in the package.
By default, all methods are selected and the filter rule is included.
Click OK when you are finished.

144 IBM Technology for Java Virtual Machine in IBM i5/OS

 See Figure 8-59.

Figure 8-59 Specify the classes and methods to monitor

13.You can repeat the previous steps to monitor any additional packages. After you have
specified all classes to profile click Finish.
14.You must be back to the properties window, as shown in Figure 8-57 on page 143. Click
OK to profile the application.
15.If you do not have the profiling and logging perspective open, you see a “confirm
perspective switch?” window open. If so, click Yes.
16.You will be in the profiling and logging perspective and see your application status as
monitoring. At the toolbar select Window → Show View → Console. This displays any
standard out data.
17.After your application completes you must see your application status reported as
terminated. Verify that all output data is correct, because you want to ensure the
application ran properly.
18.Expand the terminated link and you see options for:
– Basic Memory Analysis
– Execution Time Analysis
– Thread analysis

Chapter 8. Analyzing JVM behavior 145

 19.Right mouse click the Basic Memory Analysis option and select Open With → Memory
Statistics. See Figure 8-60.

Figure 8-60 Gather basic memory usage statistics in the profiling and logging perspective

20.You see statistics for memory and object usage, sorted by application package. You must
see these metrics, as shown in Figure 8-61:
– Total Instances
The total number of instances that have been created of the selected package, class,
or method.
– Live Instances
The number of instances of the selected package, class, or method where no garbage
collection has taken place.
– Collected
The number of instances of the selected package, class, or method that were removed
during garbage collection.
– Total Size
The total size (in bytes) of the selected package, class, or method of all instances that
were created for it, including objects that were garbage collected.
– Active Size
The summed size of all live instances.

146 IBM Technology for Java Virtual Machine in IBM i5/OS

 See Figure 8-61.

Figure 8-61 View basic memory usage statistics in the profiling and logging perspective

In this example you can see that the default package, which is where the test application
resides, generates the most activity. If you expand the default package you can see
statistics for primitives such as character, byte, integer, and so on.
If the application is still active, you can continue running it and analyze the number of
objects and memory usage. The profiling and logging perspective has icons which show
increases or decreases. You can use this technique to help detect excessive object
creation or a memory leak. See Figure 8-62.

Figure 8-62 Memory statistics view shows increases and decreases in objects

Chapter 8. Analyzing JVM behavior 147

 21.The next step is to analyze response time components. At the profiling monitor view
right-click the execution time analysis option and select Open With → Execution
Statistics. See Figure 8-63.

Figure 8-63 Gather basic response time statistics in the profiling and logging perspective

22.You must see a view with response times sorted by application package. See Figure 8-64.
You can then expand the packages and the response times for individual methods. You
must see the following metrics:
– Base Time
This is the time taken to execute the invocation, excluding the time spent in other
methods that were called during the invocation. This is typically the most useful metric.
– Average Base Time
The base time divided by the number of calls.
– Cumulative Time
This is the time taken to execute all methods called from an invocation. If an invocation
has no additional method calls, then the cumulative time is equal to the base time.
– Calls
The number of calls made by a selected method.

Figure 8-64 Response times sorted by package

In this example you can see that the mypackage package has the largest contribution to
response time.

148 IBM Technology for Java Virtual Machine in IBM i5/OS

23.If you expand the packages you can view response times for individual methods. See
Figure 8-65. You can click the Base Time (seconds) column heading (or any other
heading) to sort in ascending or descending order.

Figure 8-65 Detailed execution time for each method

In this example you can see the following:

– Total execution time is 5.13 seconds (cumulative time for the mainpgm class) as
shown in Figure 8-65.
– The main method in the mainpgm class execution time is .82 seconds (base time).
– The con_discon class execution time is 1.15 seconds, and the getConnection method
is the largest component.
– The getdtaara class execution time is 1.95 seconds and the getDtaaraparms is the
largest component.
– The runsql class execution time is 1.21 seconds and the setSqlstatement is the largest
component.
You can right-click any row and see options for viewing the source code or method
invocation.
24.The next step is to use the performance call graph to view the application execution
environment. At the profiling monitor view right-click the execution time analysis option
and select Open With → Performance Call Graph.
The performance call graph shows the threads and application method’s in a hierarchical
relationship.

Chapter 8. Analyzing JVM behavior 149

 See Figure 8-66.

Figure 8-66 Performance call graph example

In this example you can see that the most lengthy method (getdtaara.setDtaaraparms())
has the thickest line. You can also see that the application is executed in a single thread. If
you hover the mouse cursor over the individual methods, you see a pop-up window that
lists the response time metrics and source file.

Application profiling and analysis: server

It is most beneficial to do application profiling and analysis during development and testing
phases, rather than in the production phase. It is also easier to do profiling and analysis with
the Rational Software Development Platform tools, such as WebSphere Development Studio
Client for iSeries Advanced Edition, rather than server-based tools such as Performance
Explorer (PEX) in i5/OS. However, you might find it beneficial to analyze the application
running on System i platform. For example, PEX shows detailed CPU usage at the job and
thread level. PEX is also beneficial for analyzing application performance issues when the
server is running with production level workloads.

This example illustrates creating a Performance Explorer (PEX) trace for a Java application
that utilizes IBM Technology for JVM, then interpreting the results with Performance Trace
Data Visualizer (PTDV). You can see how to determine the CPU time for the JVM process,
and also the jobs that access i5/OS resources via the IBM Toolbox for Java. You can also see
how to determine the relative impact of garbage collection. PTDV can be downloaded from
the IBM alphaWorks® Web site at:
http://www.alphaworks.ibm.com/tech/ptdv

Note: i5/OS provides the support for creating a Performance Explorer trace. You must
install the Performance Tools for iSeries product (5722-PT1) if you want to print reports.

150 IBM Technology for Java Virtual Machine in IBM i5/OS

Perform the following steps:
1. Start a 5250 emulator session and sign in.
2. Enter the following command to create a PEX definition:
ADDPEXDFN DFN(TPROF5MS) TYPE(*TRACE) JOB((*ALL)) MAXSTG(1000000) INTERVAL(5)
TRCTYPE(*SLTEVT) SLTEVT(*YES) BASEVT((*PMCO *NONE *FORMAT2)) TEXT('Tprof with
sampling interval of 5 ms')
3. This command creates a trace that samples every 5 milliseconds and records specific
events. Next, start the trace by issuing the following command (give your trace a good
descriptive name):
STRPEX SSNID(JavaTprof) DFN(TPROF5MS)
Notice that you want to use the PEX definition created in the previous step: TPROF5MS.
4. Run your Java application.
5. Enter the command:
ENDPEX SSNID(*SELECT)
6. The ENDPEX screen shows how many events have been collected. Press F5 to refresh
the event count. For Java applications 100,000 to 500,000 events usually provide a good
sample. After enough events have been collected, select your collection and press Enter.
7. Start PTDV on your workstation.
8. You must see the load PEX data window open as shown in Figure 8-67.

Figure 8-67 Specify the PEX data to import

9. Select the Thick Client tab and enter the server name, user name and password.
10.You can click the Browse button to search for your PEX collection. Otherwise, enter the
library and collection name manually.
11.Click the Start Processing button to start the analysis.
12.You must see a TPROF options window open. Accept the defaults and click OK.
You must be at the cumulative information tab, which shows information about when the
trace was taken, the number of events recorded, sampling interval, and sample time. You
must also see system information such as the system model, amount of processors,
number of disk arms, and other data.

Chapter 8. Analyzing JVM behavior 151

 13.Click the Job/Thread List tab. You must see a list of events causing jobs and metrics
such as the number of events and CPU usage. Click the Cum CPU Time (ns) column to
sort it in descending order.
14.Figure 8-68 shows the jobs consuming the most CPU. In this example the WebSphere job
(TEST32JVM) consumes most of the CPU amongst all jobs (the ‘//’ job represents the
SLIC threads activity and we do not count it).

Figure 8-68 Job listing sorted by cumulative CPU time

The application performs a database query, using the Native JDBC driver. Notice that the
associated QSQSRVR jobs consumed a fairly small amount of CPU.

152 IBM Technology for Java Virtual Machine in IBM i5/OS

 15.Expand the job link and you see the job’s threads and the associated number of events
and CPU usage, as shown in Figure 8-69.

Figure 8-69 View events and CPU time by job thread

16.The next step is to view the details for the TEST32JVM job. Double-click it.

Chapter 8. Analyzing JVM behavior 153

 17.Click the Bucketized Tprof tab to get a more user friendly view of the data, as shown in
Figure 8-70.

Figure 8-70 View the event count by category

18.You must see a component level breakdown view for the number of hits to SLIC (System i
microcode), MI (i5/OS tasks), PASE (where IBM Technology for JVM runs), and JAVA.
Note that the PASE component has the most hits in this example. This is because the
applications were run with IBM Technology for JVM, and there was very little other
workload running during the trace. If they had been run with Classic JVM the hits would be
displayed in the SLIC and Java breakdown categories.
19.Expand the PASE breakdown category. See Figure 8-71.

Figure 8-71 View the activity within the PASE

20.The PASE breakdown shows the relative activity for the IBM Technology for JVM
components.

154 IBM Technology for Java Virtual Machine in IBM i5/OS

8.7 Application debugging
In most situations the Java application developer uses workstation-based tools to build and
test their applications prior to deployment on System i platform. However, there might be
situations when the application is deployed to System i platform and it does not function
properly, especially in production environments under heavy load. In this section you learn
how you can debug a Java application running on System i platform.

8.7.1 Review any available data

Your first step must be to verify that the application has in fact run properly in a test
environment (on a workstation or on System i platform). Assuming that it has, check any
available job log, spool file, or JVM console information for clues. For example, your
application might be using the IBM Toolbox for Java classes. The password for the i5/OS
user profile that it is using might be expired. If so, you would see an authentication error
message to that effect in the JVM console. Also, other issues such as network problems
might be causing functional problems in the application.

8.7.2 Debugging a Java application running on System i platform with

WebSphere Development Studio Client for iSeries
WebSphere Development Studio Client for iSeries provides a complete development
environment for traditional interactive and batch applications, Web applications, and Java
applications. It includes local test environments for testing and debugging your stand-alone,
and WebSphere Application Server based applications. However, you have to test your
application on a server (System i platform in our case) before deploying it in a production
environment.

WebSphere Development Studio Client for iSeries has a comprehensive debugging

environment for your Java applications. It allows you to set break points in your code, monitor
variables, view thread status, and many other tasks. WebSphere Development Studio Client
for iSeries allows you to do Java development on your workstation and you can test and
debug your applications locally (the application runs in a JVM on your workstation) or
remotely (the application runs in a JVM on System i platform).

Restriction: At the time of this writing, only the Classic JVM can be used for remote
debugging in WebSphere Development Studio Client for iSeries version 6.0.1. However
the approach presented in this section is applicable to IBM Technology for JVM when this
support is available.

The following are the basic steps you have to follow to debug a Java application remotely on
a System i platform:
1. Create or import a Java application project in your WebSphere Development Studio Client
for iSeries workspace.
2. Create, import, or modify the Java source files for your application.
3. Specify the appropriate breakpoints in your application code.
4. Test the application locally to ensure it actually works.
5. Define a remote system explorer connection to the System i platform.

Chapter 8. Analyzing JVM behavior 155

 6. Copy your application to the System i file system.
7. Start a debug session to run the application on the System i platform.
8. Step through the application and monitor the variables, thread status, and output data.

Steps 1 and 2 are not covered here. If necessary, refer to Rational Application Developer V6
Programming Guide, SG24-6449.

Specify appropriate breakpoints in application code

Setting breakpoints in WebSphere Development Studio Client for iSeries is easy. All you
have to do is open the source file in the editor, and double click the mouse in the left side bar,
called marker bar, of the statements you would like to step into. You can remove breakpoints
by double-clicking the breakpoint icon.

Figure 8-72 shows that an icon is placed next to the line of code where a breakpoint is set.

Figure 8-72 Setting breakpoints in WebSphere Development Studio Client for iSeries

Test the application locally

Before testing the application on the System i platform, it makes sense to ensure that the
application runs in the local workstation test environment. This is also very easy in
WebSphere Development Studio Client for iSeries. All you have to do is right-click the class
(which must have a main() method) and select Run → Java Application, as shown in
Figure 8-73.

156 IBM Technology for Java Virtual Machine in IBM i5/OS

See Figure 8-73.

Figure 8-73 Testing an application using the workstation JVM

If there is no graphical user interface, all input and output is done at the console. If there are
graphics they must open in a window on your workstation. You can check the console for any
error messages or output data as shown in Figure 8-74.

Figure 8-74 Application outputs open in the console

Chapter 8. Analyzing JVM behavior 157

 Define a remote system explorer connection to the System i platform
In the previous step you tested the application in the local workstation JVM. WebSphere
Development Studio Client for iSeries allows you to debug the application directly on System i
platform. You have to define a remote system connection to the System i platform if you want
to do this. The remote system connection provides the communications link between the
workstation development environment and the System i platform.

If you have already created a remote system explorer (RSE) connection to your System i
platform you can ignore this section. Otherwise, these are the steps you have to perform:
1. At your WebSphere Development Studio Client for iSeries workspace open the remote
system explorer perspective by selecting Window → Open Perspective → Other... →
Remote System Explorer.
2. At the top of the Remote Systems view expand New Connection.
3. Right-click iSeries... and select New connection.
If you have an existing profile it must open as the parent profile option. You can use that,
or specify a new profile, as shown in Figure 8-75.

Figure 8-75 Define a remote connection to System i platform

4. Enter a descriptive value for the location name, and the host name or TCP/IP address of
your System i platform.
5. Click Finish.
6. In the Remote Systems view right-click your server connection and select the Connect
option. Enter an appropriate user profile and password. You must be connected. You can
view library objects, the System i file systems, and other capabilities.

158 IBM Technology for Java Virtual Machine in IBM i5/OS

 See Figure 8-76.

Figure 8-76 Connected to System i platform

7. Now that you have connected to the server you can copy the application to the System i
platform.

Copy your application to the System i file system

Perform the following steps:
1. Return to the Java perspective by clicking on the Java perspective icon, as shown in
Figure 8-77.

Figure 8-77 Switching between perspectives in WDSc for i5/OS

2. Right-click your Java application project and select the Export option.
3. You must see numerous options, including file system (mapped network drive), FTP, and
remote file system (remote system explorer connection created earlier). To use the
connection created earlier, select the Remote file system option and click Next.
4. You must see all the objects in your local project selected (check boxes selected). Click
the Browse... button next to the To directory field.

Chapter 8. Analyzing JVM behavior 159

 5. Expand the link to the appropriate server and navigate to a directory on the System i
platform. In this example the project is copied to the /home directory, as shown in
Figure 8-78.

Figure 8-78 Specify a location for the Java program objects

6. Click the OK button, then Finish to copy the files to the System i file system.
7. You can use a mapped network drive or the i5/OS WRKLNK command to verify that the
Java classes and other resources have been copied to the System i file system.
Now that the objects are on the System i file system you can run the application in debug
mode.

Start a debug session on the System i platform

Perform the following steps:
1. At your WebSphere Development Studio Client for iSeries session locate your Java
project and select the application with the appropriate main() method.
2. At the main menu select Run → Debug... option.
3. At the Debug window select the Java Application in iSeries Server option. Click the
New button.
4. Specify an appropriate name for the debugging session.
5. Click the Browse... button next to the Remote iSeries working folder field.

160 IBM Technology for Java Virtual Machine in IBM i5/OS

 See Figure 8-79.

Figure 8-79 Launch the application in debug mode on System i platform

6. At the Browse for folder window expand the remote system connection link you specified
earlier and locate the folder you copied the Java application to and click OK.

Figure 8-80 Select the folder containing the Java application

Chapter 8. Analyzing JVM behavior 161

 Note: The working folder’s format is as follows:
RSEProfile.ConnectionName:directory

See Figure 8-80. Note that the Include working folder in classpath option is selected.

7. Click the Properties tab. Enter java.version for the property name and 1.5 for the value.
8. Click the Append button. This ensures that JDK 1.5 is used on System i platform. See
Figure 8-81.

Figure 8-81 Ensure that JDK 1.5 is used on System

Restriction: At the time of this writing, only Classic JVM can be used for remote
debugging on System i platform.

9. The classpath includes the working folders you uploaded to the System i platform. If you
have to specify additional classpath information or other environment variables, click the
Environment tab and specify each variable and the value. In this example the IBM
Toolbox for Java has been added.

162 IBM Technology for Java Virtual Machine in IBM i5/OS

 See Figure 8-82.

Figure 8-82 Set the CLASSPATH environment variable

10.Enter the appropriate properties, environment variables, library list, and any other
necessary parameters. Then click Apply.
11.Click the Debug button. You might be prompted to enter a user profile and password to
connect to the System i platform. Enter the necessary values and click OK.
12.You must see a progress window open, then another window asking you to switch to the
debug perspective. Click Yes.

Step through the application

Perform the following steps:
1. You must now be in the Debug perspective and see your application paused at the first
breakpoint you defined.
2. In the debug view you must see the application thread in a suspended state, awaiting you
to step through the application. In the console view you must see the i5/OS job information
as shown in Figure 8-83.

Chapter 8. Analyzing JVM behavior 163

 See Figure 8-83.

Figure 8-83 Debug session for a Java application running on System i platform

3. If you are interested in looking at the i5/OS job details, start a 5250 emulator session and
sign in. You can then use the WRKACTJOB or WRKJOB command to view the job log.
Figure 8-84 shows an example of what you must see.

Display Job Log

System: MYSERVER
Job . . : QZRCSRVS User . . : QUSER Number . . . : 378751

Job 378751/QUSER/QZRCSRVS started on 10/06/06 at 10:53:32 in subsystem

QUSRWRK in QSYS. Job entered system on 10/06/06 at 10:53:32.
Printer device PRT01 not found.
ACGDTA for 378751/QUSER/QZRCSRVS not journaled; reason 1.
Job changed successfully; however errors occurred.
User USERID from client 10.10.136.186 connected to server.
Printer device PRT01 not found.
ACGDTA for 378751/QUSER/QZRCSRVS not journaled; reason 1.
Job changed successfully; however errors occurred.
Library QSYS2924 not found.
Client request - run program QSYS/QWCRTVCA.
Client request - run program QSYS/QUSRJOBI.
Client request - run program QSYS/QWTCHGJB.
Client request - run program QSYS/QWCRTVCA.
Client request - run command 'JAVA '.
Figure 8-84 i5/OS job log for the debugging process

164 IBM Technology for Java Virtual Machine in IBM i5/OS

 The QZRCSRVS job is not the Java job, but just a job that provides a connection to the
debugger.
4. Click the Resume icon to advance to the next breakpoint you defined in the source code.
See Figure 8-85.

Figure 8-85 Use the Resume action to step through the application

5. Click the Resume icon to continue stepping through the application and determine if and
where there is an application error. You can also monitor the values of variables and
determine the status of each thread. See Figure 8-86.

Figure 8-86 View application variables and thread status

6. Figure 8-86 shows an example of the mainpgm class instantiating a JavaBean (the
getdtaara class). The getdtaara class includes a method that reads a decimal data area on
the System i platform, into a variable called “thevalue”. In this example, “thevalue” is .66.
7. You can continue clicking the Resume icon to continue stepping through the application
(or other icons, like Step over or Step into) and determine if and where there is an
application error. You can also monitor the console output.

Chapter 8. Analyzing JVM behavior 165

 See Figure 8-87.

Figure 8-87 View console output and application variables

8. Depending on how many and where the breakpoints were set, you might have to
experiment some to determine the exact cause of the application problem.
You can also use the remote debugging function to get an idea of the application response
time components.

166 IBM Technology for Java Virtual Machine in IBM i5/OS

 A

Appendix A. Running WebSphere Application

Server with IBM Technology
for JVM
This appendix describes the usage of the IBM Technology for JVM with WebSphere
Application Server on System i platform. The topics covered are:
򐂰 “System requirements for i5/OS” on page 168
򐂰 “Switching between JVMs in WebSphere Application Server 6.1” on page 168
򐂰 “Enable an existing WebSphere Application Server profile with IBM Technology for JVM”
on page 170
򐂰 “Determining the JVM in use” on page 171
򐂰 “Configuring the WebSphere Application Server JVM” on page 173
򐂰 “Monitoring the WebSphere Application Server JVM” on page 174
򐂰 “Verbose garbage collection and logging” on page 178
򐂰 “Additional information” on page 179

© Copyright IBM Corp. 2007. All rights reserved. 167

System requirements for i5/OS
WebSphere Application Server version 6.1 is supported on both Classic JVM and IBM
Technology for JVM on i5/OS V5R4M0 or later releases. Earlier versions of WebSphere
Application Server are only supported on Classic JVM. i5/OS releases prior to V5R4M0 do
not support IBM Technology for JVM. These are the basic requirements for WebSphere
Application Server version 6.1 running on V5R4M0. For a complete list of hardware and
software prerequisites refer to:
http://publib.boulder.ibm.com/infocenter/wasinfo/v6r1/topic/com.ibm.websphere.base
.iseries.doc/info/iseries/ae/cins_is_prqsvr.html

Some of the prerequisites are listed as follows:

򐂰 i5/OS V5R4M0 and cumulative program temporary fix (PTF) package C6192540 or later
򐂰 Java Developers Kit (JDK) V5.0
– 5722-JV1 option 7
– 5722-JV1 option 8
򐂰 i5/OS Qshell utilities (5722-SS1 option 30)
򐂰 i5/OS Host Servers (5722-SS1 option 12)
򐂰 i5/OS Extended Base Directory Support (5722-SS1 option 3)
򐂰 i5/OS Portable Application Solutions Environment (5722-SS1 option 33)
򐂰 WebSphere Application Server version 6.1 group PTF SF99323 version 5 or later
– DB2 for i5/OS group PTF SF99504 version 4 or later
– Java group PTF SF99291 version 3 or later
– HTTP server group PTF SF99114 version 3 or later

Important: If you install any of the listed products after installing the i5/OS cumulative
PTF package, you must reinstall the cumulative PTF package to ensure all required
PTF are applied. Reinstall the WebSphere Application Server group PTF if you installed
5722-JV1 option 7, 8, or 5722-SS1 option 33 (i5/OS PASE) after installing group PTF.

Refer to the WebSphere Application Server group PTF cover letter for installation
instructions. Ensure you run the update script, which actually applies the WebSphere
Application Server fixes.

Switching between JVMs in WebSphere Application Server 6.1

WebSphere Application Server 6.1 for i5/OS can run with Classic JVM and IBM Technology
for JVM. It includes a Qshell script than can switch between JVMs. This section covers:
򐂰 The enableJVM script that specifies the JVM for one or more WebSphere Application
Server profiles
򐂰 An example of stopping a profile, changing the JVM and restarting the profile
򐂰 How to determine which JVM is in use

168 IBM Technology for Java Virtual Machine in IBM i5/OS

The enableJVM script
By default, WebSphere Application Server version 6.1 specifies Classic JVM for all profiles
that are created.

Attention: WebSphere Application Server 6.1 supports installing the product multiple
times on the same server or logical partition. To get a list of all of your installed
environments including the default profile location and install library, you can use the
following script:
/QIBM/ProdData/WebSphere/AppServer/V61/base/bin/querywasinstalls

In these examples the default installation location is assumed.

WebSphere Application Server version 6.1 for iSeries includes an i5/OS Qshell script that can
switch the JVM type, if you do not use the default installation directory, adjust these
instructions accordingly:
/QIBM/ProdData/WebSphere/AppServer/V61/Base/bin/enableJvm

This script requires the user to have *ALLOBJ authority. The command syntax is:
enableJVM parameters

Here parameters are:

򐂰 -jvm [std32 | classic]
This is a required parameter that specifies the JVM in use:
– std32 = IBM Technology for JVM
– classic = Classic JVM
򐂰 -profile <profile_to_be_enabled>
This is an optional parameter that specifies the profile to enable.
򐂰 -verbose
This is an optional parameter that enables additional trace statements.

Important: If a profile is not specified, all profiles are switched to the specified JVM.

Here are several examples:

򐂰 enableJVM -jvm std32 -profile default
This sets the default profile to run on IBM Technology for JVM.
򐂰 enableJVM -jvm classic
This sets ALL profiles to run on Classic JVM.
򐂰 enableJVM -jvm std32 -profile default -verbose
This sets the default profile to run on IBM Technology for JVM and display additional
details on the screen.

Attention: With WebSphere Application Server Network Deployment V6.1 it is possible

to include servers with Classic JVM and IBM Technology for JVM in the same cell.

Appendix A. Running WebSphere Application Server with IBM Technology for JVM 169
Enable an existing WebSphere Application Server profile with
IBM Technology for JVM
These instructions provide a complete set of steps for switching to IBM Technology for JVM:
1. Sign in to the System i platform with an appropriate userID and password.
2. Enter STRQSH on the CL command line and hit Enter.
3. At the Qshell prompt, run the following command:
cd /QIBM/ProdData/WebSphere/AppServer/V61/Base/bin
4. If your profile is running, stop it (we use the default profile) as shown in Figure A-1:
stopServer -profileName default

QSH Command Entry

$
> cd /QIBM/ProdData/WebSphere/AppServer/V61/Base/bin
$
> stopServer -profileName default
ADMU0116I: Tool information is being logged in file

/QIBM/UserData/WebSphere/AppServer/V61/Base/profiles/default/logs/server1/stopS
erver.log
ADMU0128I: Starting tool with the default profile
ADMU3100I: Reading configuration for server: server1
ADMU3201I: Server stop request issued. Waiting for stop status.
ADMU4000I: Server server1 stop completed.
===>
F3=Exit F6=Print F9=Retrieve F12=Disconnect
F13=Clear F17=Top F18=Bottom F21=CL command entry
Figure A-1 Ending the profile

5. Invoke the enableJVM script to change your profile to use the IBM Technology for JVM:
enableJVM -profile default -jvm std32
See Figure A-2.

QSH Command Entry

> enableJVM -profile default -jvm std32

ADEJ0014I: Enabling profile default to use the specified JVM.
ADEJ0002I: Success: The profile will now use the specified JVM.
$
===>
F3=Exit F6=Print F9=Retrieve F12=Disconnect
F13=Clear F17=Top F18=Bottom F21=CL command entry
Figure A-2 Enable JVM for a specific server profile

170 IBM Technology for Java Virtual Machine in IBM i5/OS

 6. Start the WebSphere Application Server profile that you changed earlier.
startServer -profileName default
See Figure A-3.

QSH Command Entry

> startServer -profileName default

CPC1221: Job 089988/QEJBSVR/SERVER1 submitted to job queue QWASJOBQ in
library QWAS61.
CWNATV00I: Application server server1 in profile default has started and is
ready to accept connections on admin port 9060.
===>
F3=Exit F6=Print F9=Retrieve F12=Disconnect
F13=Clear F17=Top F18=Bottom F21=CL command entry
Figure A-3 Starting the profile

7. Your server must now run in IBM Technology for JVM.

Determining the JVM in use

There are two basic ways to determine which JVM is used by a WebSphere Application
Server profile:
򐂰 View the job log for your WebSphere Application Server profile.
򐂰 View the SystemOut.log file for your WebSphere Application Server profile.

The job log option is the easier one to use and is illustrated.

View the job log

To view the job log, perform the following instructions:
1. Find your active profile by using the WRKACTJOB SBS(QWAS61) command.
2. Use option 5, work with the job, then option 10, display job log, for the application server
job.
3. Press the F10 key to display the detailed messages. If you use Classic JVM, you see the
message shown in Figure A-4.

Appendix A. Running WebSphere Application Server with IBM Technology for JVM 171
 See Figure A-4.

Display Job Log

System: RCHAS60
Job . . : MYAPPSVR02 User . . : QEJBSVR Number . . . : 113772

Job 113772/QEJBSVR/MYAPPSVR02 started on 08/11/06 at 15:14:44 in subsystem

QWAS61 in QWAS61. Job entered system on 08/11/06 at 15:14:44.
ACGDTA for 113772/QEJBSVR/MYAPPSVR02 not journaled; reason 1.
Java Virtual Machine is Classic.
Server starting with user profile QEJBSVR and JDK 1.5.0.03-_13_feb_2006.
Statistics elapsed time reset.
WebSphere application server MyAppSvr02 ready.

Bottom
Press Enter to continue.

F3=Exit F5=Refresh F10=Display detailed messages F12=Cancel

F16=Job menu F24=More keys
Figure A-4 Display messages for Classic JVM

If it is IBM Technology for JVM, then you see the message shown in Figure A-5.

Display Job Log

System: RCHAS60
Job . . : MYAPPSVR02 User . . : QEJBSVR Number . . . : 113772

Job 113772/QEJBSVR/MYAPPSVR02 started on 08/11/06 at 15:14:44 in subsystem

QWAS61 in QWAS61. Job entered system on 08/11/06 at 15:14:44.
ACGDTA for 113772/QEJBSVR/MYAPPSVR02 not journaled; reason 1.

Java Virtual Machine is IBM Technology for Java. PID(343)

Server starting with user profile QEJBSVR and JDK 1.5.0.

Statistics elapsed time reset.

WebSphere application server MyAppSvr02 ready.

Bottom
Press Enter to continue.

F3=Exit F5=Refresh F10=Display detailed messages F12=Cancel

F16=Job menu F24=More keys
Figure A-5 Display messages for IBM Technology for JVM

4. Note that the process identifier PID(343) is shown in the job log. You can use this
information in case you have to do a heap dump or other diagnostics described in
Chapter 8, “Analyzing JVM behavior” on page 83.

172 IBM Technology for Java Virtual Machine in IBM i5/OS

Configuring the WebSphere Application Server JVM
Configuring the IBM Technology for JVM that is used with WebSphere Application Server is
easy if you use the administrative console:
1. At a console session click Servers → Application Servers → <application server
name>.
2. At the application servers configuration panel click Java and Process Management →
Process Definition → Java Virtual Machine. See Figure A-6.

Figure A-6 WebSphere Application Server V6.1 console makes it easy to customize JVM properties

3. In this example you can see that verbose garbage collection has been specified. If you
see no values in the Initial Heap Size and Maximum Heap Size fields, then the default
values are used:
initial heap size: 50 MB
maximum heap size: 256 MB
These are defined in the following file:
<WAS_install_root>/<edition>/classes/properties/os400j9.systemlaunch.properties
4. You can change these if you specify the appropriate parameters and click the Apply
button. You are then prompted to save or review the changes. Click the Save link after
ensuring all parameters have been entered correctly.
5. Note that the generic JVM arguments parameter is blank in Figure A-6. This tells you that
the default garbage collection policy (optimum throughput) is in place.

Appendix A. Running WebSphere Application Server with IBM Technology for JVM 173
 These are some commonly used JVM arguments for the WebSphere Application Server
version 6.1 environment:
– -agentlib:QWASJVMTI (enables JVM profiling)
– -Xgcpolicy:optthruput (specifies the optimum throughput garbage collection policy)
– -Xgcpolicy:optavgpause (specifies the minimum average pause garbage collection
policy)
– -Xgcpolicy:gencon (specifies the generational garbage collection policy)
– -Xgcpolicy:subpool (specifies the subpool garbage collection policy)
Because IBM Technology for JVM uses the just-in-time (JIT) compiler exclusively, you do
not specify the -Djava.compiler=jitc parameter like you would with Classic JVM.

Attention: Be very careful when you specify JVM command line arguments. Mistakes
as simple as case sensitivity in the parameters can prevent the JVM from starting. JVM
command line arguments are specified in the profile’s server.xml file.

JVM parameters exclusive to Classic JVM may prevent the IBM Technology for JVM
from starting, and vice versa.

Important: At the time of this writing, the JVM profiling functions were not working.

6. Restart your WebSphere profile to enable new parameters.

Monitoring the WebSphere Application Server JVM

The measurement and analysis tools discussed in Chapter 8, “Analyzing JVM behavior” on
page 83 can be utilized with a WebSphere Application Server profile that utilizes IBM
Technology for JVM.

Monitoring with Tivoli Performance Viewer

This section illustrates how you can use the Tivoli Performance Viewer, which is integrated in
the WebSphere Application Server administrative console, to monitor JVM performance.
These concepts are applicable to IBM Technology for JVM and Classic JVM:
1. At a console session click Monitoring and Tuning → Performance Monitoring
Infrastructure (PMI) → <application server name>.
2. At the Performance Monitoring Infrastructure (PMI) panel you must be in the configuration
tab. This specifies the monitoring settings when the application server starts. The default
settings are PMI enabled, with the basic statistic set.

174 IBM Technology for Java Virtual Machine in IBM i5/OS

 See Figure A-7.

Figure A-7 Configuring PMI

3. Click the Runtime tab. This enables you to dynamically specify the performance
monitoring policies.
4. Expand the basic monitoring statistic link and scroll down to the JVM options. You must
see the following JVM metrics available for monitoring:
– JVM Runtime.ProcessCpuUsage (the percentage of CPU usage of the JVM runtime)
– JVM Runtime.UpTime (the total time, in seconds, that the JVM has been running)
– JVM Runtime.UsedMemory (the used memory, in KB, of the JVM runtime
– JVM Runtime.HeapSize (the total memory, in KB, of the JVM runtime)
These metrics are available if JVM profiling is enabled:
– Garbage Collection.GCTime
– Garbage Collection.GCIntervalTime
– Garbage Collection.GCCount
– Object.ObjectFreedCount
– Object.ObjectAllocateCount
– Object.ObjectMovedCount
– Thread.ThreadEndedCount
– Thread.ThreadStartedCount
– Monitor.WaitForLockTime
– Monitor.WaitsForLockCount
5. At your console session click Performance Viewer → Current Activity → <application
server name>.

Appendix A. Running WebSphere Application Server with IBM Technology for JVM 175
 6. Expand the Performance Modules link and click the JVM Runtime box.
See Figure A-8.

Figure A-8 Select the JVM runtime statistics for viewing

7. Click the View Module(s) button to see the JVM runtime data.
8. You must see the JVM statistics in either tabular or graphic format. The following are the
requirements for viewing graphics:
– The Adobe Scalable Vector Graphics plugin for your browser
– i5/OS PASE (required for IBM Technology for JVM)
– Set the following parameters:
-Djava.awt.headless=true
-Dos400.awt.native=true
– In the following file:
os400j9.systemlaunch.properties
This file is located in:
<WAS_install_root>/<edition>/classes/properties/os400j9.systemlaunch.propert
ies

176 IBM Technology for Java Virtual Machine in IBM i5/OS

 See Figure A-9.

Figure A-9 View a graphical representation of the JVM runtime

9. In Figure A-9 you can also see that the total heap size is relatively constant and that CPU
usage is minimal.
10.Figure A-9 shows an example of the JVM runtime data in graphical format. In this case the
graph scaling was modified to improve the view in order to see how the used memory
changes.
11.Figure A-10 shows an example of the JVM runtime data in tabular format.

Figure A-10 View a tabular representation of the JVM runtime

Appendix A. Running WebSphere Application Server with IBM Technology for JVM 177
 12.You can use the Tivoli Performance Viewer to get a conceptual view of your JVM’s
memory usage. If the memory usage is fairly consistent over time, it is a good indicator
that currently there are no memory leaks in your applications.
On the other hand, if memory usage continues to grow, even with a steady workload, it
might indicate a memory leak and you have to utilize the techniques in Chapter 8,
“Analyzing JVM behavior” on page 83 to analyze the problem in more detail.

Verbose garbage collection and logging

You can enable verbose garbage collection for IBM Technology for JVM or Classic JVM
using the WebSphere Application Server administrative console and selecting the Verbose
garbage collection check box as indicated in Figure A-6 on page 173. By default verbose
garbage collection statistics is written to the following directory:
<WAS_install_root>/<edition>/profiles/<profile>/logs/<profile>

In this directory the verbose garbage collection statistics is written to the following files:
򐂰 For IBM Technology for JVM: native_stderr.log
򐂰 For Classic JVM: native_stdout.log

The SystemOut.log file in the aforementioned directory includes helpful JVM information. The
excerpt in Example A-1 shows that the IBM Technology for JVM is used.

Example: A-1 Excerpt showing the use of IBM Technology for JVM
WebSphere Platform 6.1 [BASE 6.1.0.1 cf10631.18] running with process name
RCHAS60_j9res_GW\RCHAS60_j9res_GW\j9res_GW and process id 212236/QEJBSVR/J9RES_GW
Host Operating System is OS/400, version V5R4M0
Java version = J2RE 1.5.0 IBM J9 2.3 OS400 ppc-32 (JIT enabled)
J9VM - 20060501_06428_bHdSMR
JIT - 20060428_1800_r8
GC - 20060501_AA, Java Compiler = j9jit23, Java VM name = IBM J9 VM

The excerpt in Example A-2 shows the heap monitor status during the server startup, in
addition to the i5/OS memory pool details.

Example: A-2 Excerpt showing heap monitor status during server startup
Heap Monitor started for 212236/QEJBSVR/J9RES_GW in subsystem QWAS61
in Pool *BASE pool ID=2
Poolsize(MB)=25107 Reserved(MB)=2
Heap total(MB)=50 Free(MB)=19 UsedHeap(MB)=30
MaxHeap(MB)=256
InitHeap(MB)=50

Example A-3 is an example of the verbose garbage collection data.

Example: A-3 Verbose garbage collection data

<af type="tenured" id="34" timestamp="Mon Oct 09 15:14:49 2006"
intervalms="3510.159">
<minimum requested_bytes="24" />
<time exclusiveaccessms="0.092" />
<tenured freebytes="368128" totalbytes="146692096" percent="0" >
<soa freebytes="0" totalbytes="146323968" percent="0" />

178 IBM Technology for Java Virtual Machine in IBM i5/OS

 <loa freebytes="368128" totalbytes="368128" percent="100" />
</tenured>
<gc type="global" id="34" totalid="34" intervalms="3531.284">
<refs_cleared soft="8" weak="3" phantom="12" />
<finalization objectsqueued="168" />
<timesms mark="234.412" sweep="7.486" compact="0.000" total="242.215" />
<tenured freebytes="49852368" totalbytes="146692096" percent="33" >
<soa freebytes="49631184" totalbytes="146470912" percent="33" />
<loa freebytes="221184" totalbytes="221184" percent="100" />
</tenured>
</gc>
<tenured freebytes="49851672" totalbytes="146692096" percent="33" >
<soa freebytes="49630488" totalbytes="146470912" percent="33" />
<loa freebytes="221184" totalbytes="221184" percent="100" />
</tenured>
<time totalms="247.748" />
</af>

Tip: It is recommended that you collect the verbose GC output to a separate file using the
-Xverbosegclog parameter in the Generic JVM arguments field. See Figure A-6 on
page 173. For further information, refer to 5.2.3, “Verbose GC output” on page 44.

Additional information
The WebSphere Application Server information center also has helpful information. It is
available at:
http://publib.boulder.ibm.com/infocenter/wasinfo/v6r1/topic/com.ibm.websphere.base
.iseries.doc/info/welcome_base.html

Appendix A. Running WebSphere Application Server with IBM Technology for JVM 179
180 IBM Technology for Java Virtual Machine in IBM i5/OS
 B

Appendix B. IBM Support Assistant

This appendix discusses IBM Support Assistant (ISA) version 3.0.1. ISA is a free, local
serviceability workbench that helps users resolve questions and problems with IBM software
products. A main goal of ISA is to become a single conduit for IBM customers to retrieve and
use serviceability tooling and also provide the core foundation of functions required for
self-help. Its features include serviceability tools, concurrent searching, quick access to
support-related information, and electronic problem management including automated data
collection. All of these functions are delivered together in the IBM Support Assistant
application to better enable users for self-help problem determination.

The focus of this appendix is in demonstrating the typical tasks for downloading, installing,
and using the tool. For more information visit the ISA Web site at:
http://www.ibm.com/software/support/isa

© Copyright IBM Corp. 2007. All rights reserved. 181

Overview of IBM Support Assistant
IBM Support Assistant is a free software product intended to help IBM customers to be more
productive with IBM products by resolving problems faster.

Important: ISA is the strategic delivery mechanism for IBM Java virtual machine (JVM)
tools.

You can use ISA when experiencing a problem. ISA offers resources for self-help that can
enable you to identify, assess, and overcome questions or problems without having to contact
IBM. When it is necessary to contact IBM, ISA offers resources for fast submission of problem
reports and immediate, automated collection of diagnostic data that can accelerate problem
resolution.

Install IBM Support Assistant

ISA 3.0.1 is available as free download from the following site:
http://www.ibm.com/software/support/isa/

Updates to the ISA application are delivered via the Updater function within ISA (refer to the
documentation delivered with the product).

Prerequisites
The following list describes the operating systems that are supported by ISA:
򐂰 Microsoft® Windows® XP SP1, 2000, and 2003 server
򐂰 Linux RedHat Advanced Server 3, Linux SuSE 9.0
򐂰 HP/UX 11
򐂰 Solaris 9
򐂰 AIX 5.2 and 5.3

Table B-1 describes Web browsers that are supported by ISA.

Table B-1 Supported Web browsers

Platform Web browser

Windows MS Internet Explorer® 6.x

Unix Mozilla 1.7 and later

Mozilla Firefox 1.0.7+

Linux Mozilla 1.7 and later

Mozilla Firefox 1.0.7+

ISA, along with the tools described in this IBM Redbook, requires a minimum of 200 MB of
free space for installation.

You can use ISA with i5/OS to gather product information, run problem determination tools,
and analyze problems, but you must install ISA on a supported workstation that you can
connect to your i5/OS. It is also recommended setting up a integrated file system share to
easily access i5/OS files from your workstation.

182 IBM Technology for Java Virtual Machine in IBM i5/OS

Downloading IBM Support Assistant from Web
There are two major components to downloading ISA from the Web. You must first download
the product itself. Then, you must run the Updater program in ISA to download any tools or
features for products you are interested in.

The following steps document how to download ISA to the Windows platform. Modify the
steps as required for your operating system. Perform the following steps:
1. Open the following URL in your Web browser:
https://www14.software.ibm.com/webapp/iwm/web/preLogin.do?source=isa
2. Click Sign in to sign in to the Web site using your IBM ID and password. If you do not
have one, register by clicking the register now link.
3. Select the latest version for your operating system and click Continue. See Figure B-1.

Figure B-1 Selecting the ISA version

4. View the license agreement and select the I agree check box to accept it.
5. Click I confirm.
6. Select the check box next to your operating system’s version and click Download now.
7. Click Save in the dialog box to save the zip file into a temporary directory on your
workstation.
8. When the download of the zip file has completed, unzip the zip files using a zip
compression utility (PKZIP, WinZip, or InfoZip) to a temporary directory, for example
C:\Downloads\isa_v3.

Note: Do not unzip the files into a directory name that contains spaces.

9. After it is extracted, run the executable file (setupwin32.exe) to start the installation.

Appendix B. IBM Support Assistant 183

 Note: There is a way to install ISA using console or silent mode, but GUI installation is
the easiest way for installing ISA.

Refer to the following for detailed information about silent installation:

<installation_dir>\Installation_and_Troubleshooting_Guide\Installation_and_T
roubleshooting_Guide.html

10.Click Next to install ISA to your workstation.

11.View and accept the license agreement, and click Next.
12.Specify an installation directory, and click Next.
13.Select the features for ISA you want to install, and click Next.
14.Review the Summary panel, and click Install.
15.When the installation completes successfully, click Finish.
16.You must now configure ISA to work with one of the available tools or features. Refer to
“Configuring and updating IBM Support Assistant” on page 184.

Configuring and updating IBM Support Assistant

After you have successfully installed ISA you must also download one or more features for
the products that you want to use. In our example we have to install WebSphere Application
Server V6.1 features to access available JVM tools.

To install a WebSphere Application Server feature, perform the following steps:

1. Start ISA. Refer to “Starting IBM Support Assistant” on page 186 for more information.
2. Click Updater.
3. Click the New Products and Tools tab as shown in Figure B-2.
4. The plug-in features are categorized by product family. Expand WebSphere, and select
WebSphere Application Server V6.1 as shown in Figure B-2.

Figure B-2 Installing WebSphere Application Server 6.1 feature

184 IBM Technology for Java Virtual Machine in IBM i5/OS

Adding the JVM tools
Next you add the JVM tools:
1. Scroll down the list of available features and expand Common Component Tools as
shown in Figure B-3.
2. Select a check box next to the tools that you want to install and click Install.

Figure B-3 Adding the JVM tools

3. A new pop-up window is displayed. Be sure to read the license and description for each
feature and tool. If you agree with the license, click I agree.
4. The installation starts. When all selected tools and features are installed, a pop-up window
opens. Click OK to restart ISA.

At the time of publication, the following features were available for use with WebSphere
Application Server:
򐂰 IBM Guided Activity Assistant
򐂰 IBM Pattern Modeling and Analysis Tool for Java Garbage Collector (PMAT)
򐂰 Memory Dump Diagnostic for Java (MDD4J)
򐂰 ThreadAnalyzer

Keeping ISA current is critical to your success in finding the most value out of the ISA tool.
Often times, new tools or features are released on an intermittent basis.

Important: It is important to occasionally check for new updates for the installed tools or to
see what other tools for your products have been made available since your installation or
last update.

Appendix B. IBM Support Assistant 185

Uninstall IBM Support Assistant
You can uninstall ISA in several ways:
򐂰 (Windows only) Use the Add/Remove Programs task in the control panel to remove the
IBM Support Assistant product.
򐂰 (Windows only) Navigate to Start → Programs → IBM Support Assistant → Uninstall
IBM Support Assistant to start the uninstallation process.
򐂰 (All platforms) Navigate to your ISA install location, double-click the _uninst folder, and
run the uninstall executable file.

Note: You can also uninstall ISA in the console or silent mode. Run the following
command from a command prompt:
<install_root>\_uninst\<uninstaller_executable_file> -console

Or:
<install_root>\_uninst\<uninstaller_executable_file> -silent -options
options.txt

For example, in Windows:

"c:\Program Files\IBM\IBM Support Assistant v3\_uninst\uninstall.exe"
-console

Use IBM Support Assistant

Use ISA to search for support information, find product information, analyze product problems
using specialized tools, and submit problem reports to IBM that are expedited with automatic
data collection.

You can also use the ISA interface to check for new IBM products and tools that are
developed for use with ISA.

Starting IBM Support Assistant

On the Windows platforms, ISA starts up with its own GUI windows. On all other platforms,
ISA launches the default browser for the system and the ISA application is accessed via the
browser interface.

To start ISA on Windows, click Start → All Programs → IBM Support Assistant → IBM
Support Assistant V3.

To start ISA on other supported platforms, perform the following steps:

1. Open the command prompt.
2. Specify startisa.sh on the command line. A Mozilla browser opens with the IBM Support
Assistant V3 welcome page.

Note: If you do not have the Mozilla browser installed, you have to manually open the
ISA URL in another supported Web browser. The URL used to access ISA is also
written to the console window where you ran the startisa.sh script. The URL is also
listed in the <install_root>/workspace/logs/isaurl.log file. Open this file, copy URL, and
paste it in a Web browser window.

186 IBM Technology for Java Virtual Machine in IBM i5/OS

Starting JVM tools
In order to access any JVM tool, you have to open the WebSphere Application Server V6.1
feature. Follow these instructions to start a JVM tool:
1. In the ISA window click Tools in the toolbar.
2. In the left menu click WebSphere Application Server 6.1.
3. You must see the list of the installed tools in the right frame. See Figure B-4.

Figure B-4 Available JVM tools

4. Click a link for a desirable tool. In our example we click Memory Dump Diagnostic for
Java (MDD4J). If you see any warning message, read its text and click OK. Some tools
have prerequisites.
5. A new window opens. Now you can use the tool. Refer to Chapter 8, “Analyzing JVM
behavior” on page 83 for the instructions about how to use each tool.

Submitting a problem report to IBM

ISA provides a service feature with an automated system and symptom-based collector. The
system collector gathers general information from your operating system, registry, and so on.
The symptom-based collection provides the unique ability to collect specific information
relating to a particular problem that you are having. It can also provide you with the ability to
automatically enable tracing that is helpful to IBM support as part of the data gathering
process.

You also can submit problems directly to IBM by creating a problem report and attaching a
generated collector file at the same time.

Appendix B. IBM Support Assistant 187

 Note: You must use the Portable Collector option to export a .jar file to your i5/OS. The
option for Collect Data is not supported on i5/OS.

Perform the following steps to collect data from your i5/OS and submit a problem report with
the collected data:
1. Start ISA. Refer to “Starting IBM Support Assistant” on page 186 for more information.
2. Click Service.
3. Click Create Portable Collector.
4. Select System collector in the Select a product box.
5. Enter the integrated file system path to your i5/OS in the Output directory field. For
example, as follows:
U:\home\collector_output
Here U: is a mapped network drive to your system’s Integrated File System (IFS) share.
6. Enter the name of the .jar file to be created in the Output file name (*.jar) field. You must
enter the .jar extension. For example, as follows:
problem.jar
7. Click Export. See Figure B-5.

Figure B-5 Creating a portable collector

8. Open an IBM Personal Communications screen and log in to your system.

9. Enter STRQSH on the CL command line and hit Enter.

188 IBM Technology for Java Virtual Machine in IBM i5/OS

 10.On the QShell command line enter the change directory command to change to your .jar
file location. For example:
cd /home/collector_output
11.Enter the following command on the Qshell command line:
jar -xvf problem.jar
Files are extracted to your directory location.
12.Enter ENV on the Qshell command line to check whether you have a JAVA_HOME
variable defined. ISA works best with Java Development Kit (JDK) 1.4 on i5/OS.
13.If you do not have a JAVA_HOME variable defined, enter the following command on the
Qshell command line:
export JAVA_HOME=/QIBM/ProdData/Java400/JDK14
You can verify the JAVA_HOME variable value by running the java -version command
on the Qshell command line.
14.Enter chmod -R 755 `find . -name '*.sh'` on the Qshell command line to make sure
your Qshell scripts have execute permission.
15.Enter ./startcollector_iseries.sh on the Qshell command line to begin system data
collection.
16.At the prompt, enter an output file name and path for the newly-created ZIP file. For
example:
/home/collector_output/12345.724.888.problemoutput.zip
17.Enter 1 at the prompt to collect system output.
18.The tool prompts you if you want to run another collection. If you are done, type No and hit
Enter.
You now have a ZIP file on your i5/OS in the location you specified. You must now upload
it to IBM using ISA.
19.Go back to ISA, and click Log into ESR.
20.Enter your IBM ID, Password, IBM customer number, and Country/region of support
contract, and click Login.
21.On the Submit Problem Report screen, identify your product, select your component,
select your contract, provide your relevant information, and attach the ZIP file that was
created with the Portable collector tool.

Troubleshooting IBM Support Assistant

There are several log files available in ISA to help with troubleshooting ISA problems.

There is an installation and uninstallation log file available in the location where ISA is
installed. The file name is log.txt.

There are log files available for problems starting ISA in the <install_root>/workspace/logs
directory. The isa_*.log files represent a set of rolling log files, where the most recent log file
is always named isa_0.log.

There is also a log file in the <install_root>/workspace/.metadata directory that contains

logging messages that might occur either before ISA logging is initialized or if logging
initialization fails.

Appendix B. IBM Support Assistant 189

 Refer to the following file for some troubleshooting information:
<ISA_installation_files_dir>\Installation_and_Troubleshooting_Guide\en\Installatio
n_and_Troubleshooting_Guide_en.html

This file is part of the directory that is created when you expand the ZIP file downloaded from
the Web. Refer to “Downloading IBM Support Assistant from Web” on page 183.

190 IBM Technology for Java Virtual Machine in IBM i5/OS

 C

Appendix C. Diagnostic Tools and Framework

for Java Dumps
This section provides information about the Diagnostic Tools and Framework for Java Dumps
(DTFJ), which has been enhanced in version 5.0 of IBM Technology for Java virtual machine
(JVM). DTFJ is a flexible framework that allows diagnostics tools to be written in a platform,
runtime, and version-independent manner.

This appendix covers the “Overview of DTFJ” on page 192.

© Copyright IBM Corp. 2007. All rights reserved. 191

Overview of DTFJ
DTFJ has two main abstract entities at its foundation, an image and runtime environment. An
image is any complete operating system image, such as a system dump file. A runtime
environment is any language-specific runtime environment, such as a JVM. DTFJ provides a
layer of abstraction, making the image format and internal data structures of the runtime
environment transparent. Because of this abstraction, DTFJ can be extended to encompass
other operating system images and languages, although we focus on the JVM in this section.
DTFJ could be extended to handle diagnostic data from other vendors’ JVMs in the future
also.

One of the core components of DTFJ is an application programming interface (API) to provide
access to data stored in the operating system image, making it simpler to write Java
diagnostics tools. In the implementation of the DTFJ API shipped with IBM Technology for
JVM, the basic abstract concepts of an image and runtime environment have been extended
by adding interfaces for Java-specific entities such as Java threads.

Tools that use the DTFJ API can gain access to the huge array of information about a Java
process that is available in a system dump. No knowledge of the system dump format or how
Java objects/structures are laid out in memory is required by a tool-writer.

Using the DTFJ API, cross-platform tools can be written in Java to process the wide range of
data available in system dump files. This includes information about the platform on which the
process is running, including:
򐂰 Physical memory
򐂰 CPU number and type
򐂰 Libraries
򐂰 Commands
򐂰 Thread stacks
򐂰 Registers

The API can also provide information about the state of the Java runtime and the Java
application it is running, including:
򐂰 Class loaders
򐂰 Threads
򐂰 Monitors
򐂰 Heaps
򐂰 Objects
򐂰 Java threads
򐂰 Methods
򐂰 Compiled code
򐂰 Fields and their values

Because you can create system dump files non-destructively (for example, using the -Xdump
command-line option or calling the static method com.ibm.jvm.Dump.SystemDump()), you
can create tools to perform analysis of system dumps collected while the JVM in question is
still running, in addition to traditional analysis of system dumps generated due to a JVM
crash.

jextract
Due to the differences in system dump format between platforms and Java releases, the
jextract utility must run against a system dump file before it can be processed using any tool
that utilizes the DTFJ API methods. jextract must be run on the system that produced the
system dump. jextract handles the JVM-version-specific and platform-specific details for you,

192 IBM Technology for Java Virtual Machine in IBM i5/OS

 producing metadata from the system dump that allows the internal structure of the JVM to be
analyzed. The output from jextract can be passed to a tool that uses the DTFJ API.

jextract is supplied with IBM Technology for JVM in the following directory:
/QOpenSys/QIBM/ProdData/JavaVM/jdk50/32bit/jre/bin

You supply the name of the input system dump file as a command-line argument to jextract.
For example, issuing the following command in a Qshell session:
jextract /home/rumsbya/core.20061002.140109.9288.dmp

This command runs jextract on the system dump file core.20061002.140109.9288.dmp.

Because no output filename is specified, jextract simply adds a .zip extension to the input
filename when creating the output file. The output zip file or files contained in it are then
typically passed as input to tools built on the DTFJ API.

Tools based on DTFJ API

The DTFJ API is intended to facilitate the writing of tools to analyze the variety of dump files
that the JVM can create. Such tools are anticipated to be developed by both IBM and
interested parties who require a customized tool for their specific requirements. Because the
DTFJ itself is relatively new, there are a limited number of tools available at the time of writing
that exploit the API. However, the following tools available through the IBM Support Assistant
use or shortly are going to use the DTFJ API:
򐂰 ThreadAnalyzer
򐂰 MDD4J
򐂰 EVTK

The DTFJ allows speedier development of further diagnostics tools due to the common API
and planned re-use of some common tool components. It is worth mentioning that a tool’s
usage of the DTFJ API is transparent to end users of the tool. Of course, as improvements to
the diagnostics tools are made and new features are implemented, undoubtedly the tool
interfaces are going to be updated to reflect these.

DTFJ API documentation

The DTFJ API documentation is shipped as a zip file with IBM Technology for JVM on most
platforms. i5/OS users can obtain this zip file by visiting the FTP site for this IBM Redbook.
Refer to Appendix E, “Additional material” on page 199 for more information about accessing
the FTP site. The DTFJ API documentation is available in the DTFJ folder of the extracted
apidoc.zip file.

For guidance on how to use the DTFJ API for tool development, refer to the following IBM
developerWorks article:
http://www-128.ibm.com/developerworks/java/library/j-ibmjava5/

We also suggest consulting the IBM Developer Kit and Runtime Environment, Java 2
Technology Edition, Version 5.0 Diagnostics Guide for more information about DTFJ:
http://www-128.ibm.com/developerworks/java/jdk/diagnosis/

Appendix C. Diagnostic Tools and Framework for Java Dumps 193

DTFJ API or JVMTI?
For tool writers, the DTFJ API presents another option for accessing information about the
state of a JVM. There naturally may arise a question as to what distinguishes the DTFJ API
available in IBM Technology for JVM from tooling mechanisms provided in the Java 2 software
development kit (SDK), such as the JVM Tools Interface (JVMTI).

The DTFJ API and JVMTI have different goals and implementations. Therefore, although
there is a degree of crossover in terms of the data that can be obtained through these
mechanisms, the way in which the data is gathered and used is markedly different in most
cases. It is not intended in any way for the DTFJ API to replace the JVMTI because they are
intended for different purposes.

Table C-1 presents the major differences between the DTFJ API and JVMTI.

Table C-1 Differences between the DTFJ API and JVMTI

DTFJ API JVMTI

Interface type Java based on extendable API Native code, based on JVM
callbacks

Typical use Offline analysis of system Live monitoring and control of

dumps; troubleshooting JVM; profiling an application

Tool format Java application running in Native library (agent) specified

separate JVM on Java command line

Tool portability Java code can be run on any Native code has to be
compatible JVM on any recompiled for different
platform platforms

Provides means to instrument No Yes

application?

Live or post-mortem analysis? Tool can run on system dump Tool runs while the JVM being
gathered non-destructively analyzed is live
from live JVM or generated
automatically in the event of
JVM failure

Impact on JVM being analyzed For post-mortem analysis, Variable. Depends on what type
none. Taking a non-destructive of events the agent has asked
system dump can take time. to be notified about
The tool runs in a separate
JVM, therefore, no impact after
the system dump is generated

Maturity Since IBM JDK 1.4.2 Since J2SE 5.0. Predecessor

JVMPI present since Java 1.1

Applicability Unique to IBM JVMs at present JSR-163 compliant, therefore,

supported across all J2SE
5.0-compliant JVMs, including
IBM Technology for JVM

194 IBM Technology for Java Virtual Machine in IBM i5/OS

 D

Appendix D. jconsole
jconsole is a standard tool to monitor runtime parameters of the Java virtual machine (JVM).
This appendix demonstrates how to start jconsole from your workstation. During the time of
writing this book jconsole, however, is not supported in WebSphere Application Server.

© Copyright IBM Corp. 2007. All rights reserved. 195

Monitoring IBM Technology for JVM remotely with a GUI
This section shows you how to use the Java Management Extensions (JMX™) utility to
monitor the IBM Technology for JVM. For this example we use the jconsole utility that is
included with J2SE 5.0 software development kit (SDK).

There are two basic steps involved:

1. Initialize the JVM that is going to be monitored.
For demonstration purposes this example has Secure Sockets Layer (SSL) and
authentication disabled. For details about using security with JMX, see the following link:
http://java.sun.com/j2se/1.5.0/docs/guide/management/agent.html#remote
We add the following command line arguments to our java invocation:
-Dcom.sun.management.jmxremote.port=44444
-Dcom.sun.management.jmxremote.authenticate=false
-Dcom.sun.management.jmxremote.ssl=false
For example:
java -Dcom.sun.management.jmxremote.port=44444
-Dcom.sun.management.jmxremote.authenticate=false
-Dcom.sun.management.jmxremote.ssl=false -cp ../jars/bughunt.jar
bughunt.hanger.Eatery
2. Connect the jconsole utility to the running process.
Find the jconsole utility on your workstation. It must be in <JDK_INSTALL>\bin, where
JDK_INSTALL is the directory to which the JDK was installed. For example:
C:\Program Files\Java\jdk1.5.0_06\bin
Open a Microsoft Windows command prompt and run the command similar to the
following:
C:\Program Files\IBM\Java50\bin>jconsole <system_hostname>:44444
Here <system_hostname> is a host name of the server with active JVM that you have
started in the previous step. The window shown in Figure D-1 must open.

Figure D-1 Monitoring and Management console (jconsole)

196 IBM Technology for Java Virtual Machine in IBM i5/OS

 Each tab has some useful information. Figure D-2 shows an example of what you see on
the threads tab.

Figure D-2 The threads tab

Explore other parameters by selecting other tabs.

Appendix D. jconsole 197

198 IBM Technology for Java Virtual Machine in IBM i5/OS
 E

Appendix E. Additional material

This IBM Redbook refers to additional material that can be downloaded from the Internet as
described below.

Locating the Web material

The Web material associated with this IBM Redbook is available in softcopy on the Internet
from the IBM Redbooks Web server. Point your Web browser to:
ftp://www.redbooks.ibm.com/redbooks/SG247353

Alternatively, you can go to the IBM Redbooks Web site at:

ibm.com/redbooks

Select the Additional materials and open the directory that corresponds with the IBM
Redbook form number, SG247353.

Using the Web material

The additional Web material that accompanies this IBM Redbook includes the following files:
File name Description
apidoc.zip This file contains Javadoc™ for several APIs

How to use the Web material

Create a subdirectory (folder) on your workstation, and unzip the contents of the Web
material zip file into this folder. Switch to any subfolder under apidoc folder, for example
DTFJ, and open the index.html file from this folder in a Web browser.

© Copyright IBM Corp. 2007. All rights reserved. 199

200 IBM Technology for Java Virtual Machine in IBM i5/OS
Related publications

The publications listed in this section are considered particularly suitable for a more detailed
discussion of the topics covered in this IBM Redbook.

IBM Redbooks
For information about ordering these publications, see “How to get IBM Redbooks” on
page 201. Note that some of the documents referenced here may be available in softcopy
only.
򐂰 WebSphere Application Server V6 Problem Determination for Distributed Platforms,
SG24-6798

Online resources
These Web sites are also relevant as further information sources:
򐂰 IBM eServer iSeries Information Center
http://publib.boulder.ibm.com/iseries/
򐂰 IBM(R) Developer Kit and Runtime Environment, Java(TM) 2 Technology Edition, version
5.0, Diagnostics Guide:
http://publib.boulder.ibm.com/infocenter/javasdk/v5r0/index.jsp
򐂰 IBM Systems Workload Estimator
http://www.ibm.com/systems/support/tools/estimator/index.html
򐂰 Java Native Interface documentation
http://java.sun.com/j2se/1.5.0/docs/guide/jni/index.html
򐂰 Performance Management for IBM System i
http://www-03.ibm.com/servers/eserver/iseries/perfmgmt/resource.html
򐂰 System i group PTFs by release
http://www-912.ibm.com/s_dir/sline003.NSF/GroupPTFs?OpenView&view=GroupPTFs
򐂰 WebSphere Application Server V6.1 Information Center
http://publib.boulder.ibm.com/infocenter/wasinfo/v6r1/index.jsp

How to get IBM Redbooks

You can search for, view, or download IBM Redbooks, IBM Redpapers, Hints and Tips, draft
publications, and additional materials, and also order hardcopy IBM Redbooks or CD-ROMs,
at the following Web site:
ibm.com/redbooks

© Copyright IBM Corp. 2007. All rights reserved. 201

Help from IBM
IBM Support and downloads
ibm.com/support

IBM Global Services

ibm.com/services

202 IBM Technology for Java Virtual Machine in IBM i5/OS

Index

Symbols D
data queue 75, 128
.class 4
dead objects 39–40
.class file 50, 52, 61
deadlock 91–92, 94, 97, 99
.java 2
debugging
.profile 24
application 97
default garbage collection policy 38, 47
Numerics default heap size 42
32-bit address space 37 defineClass() 64
5722-JC1 10 defining heap 111
5722-JV1 10 Diagnostic Tooling Framework for Java (DTFJ)
64-bit PASE native methods 76 API 193–194
API documentation 193
overview 192
A direct execution program 11
activity levels 129 distributed program call 75
ADDENVVAR 25 DMPJVM 29
address space 80 DSPJVAPGM 14
adopted authority 77 dynamic linking 50
allocate space 40
allocation
failure 45 E
request 45 enableJVM 168–169
application EVTK tool 138, 140, 193
performance 99, 141 overview 133
startup 52 Extensible Program Call Markup Language 75
asynchronous garbage collection 12 extension
class path 50
classloader 50, 54, 67
B
basic configuration 23
bootstrap classloader 50, 54, 67 F
bytecode 51 findHelperForClassLoader() 63
modification 53, 68 finding dependencies 32
findSharedClass() 64
fixed-size heaps 43
C free list 41
C function 76
C programming language 74–75
C++ 75 G
C++ programming language 74 garbage collection 6, 35–36, 41, 46
call stack 98 "Stop The World" 38–39
class loading 49 available options 37
class path 50 choosing a policy 47
classes 50 collection cycles 36
Classic JVM 35, 37, 42, 52, 91, 110 configuring 30
classloader 50, 61 default GC policy 47
cache 50 fine-tuning operation 37
delegation model 50, 54 free list 41
classpath 58 garbage collector threads 39
code reuse 2 identifying causes of poor GC performance 44
command-line options 25 in Classic JVM 36
common problems 25 interpreting verbose GC output 44
compaction 36 live object tracing 38
compiler 2 mark phase 36, 38–39, 46
CRTJVAPGM 14 marking live objects 38

© Copyright IBM Corp. 2007. All rights reserved. 203

 pause time 38–40, 47 IBM pattern modeling and analysis tool
policies 37 Java garbage collector 185
policy 35 IBM persistent reusable JVM 52
shared classes and garbage collection 63 IBM POWER 128
sweep phase 36, 38–39, 46 IBM support
sweeping dead objects 39 collecting MustGather data 91
tuning options in Classic JVM 36 IBM Support Assistant 94, 181, 193
verbose GC output 37, 44, 46–47 configure 182
garbage collection (GC) 6 install 182
gencon 40, 46 overview 182
flipping allocate and survivor spaces 40 portable collector 188
generational concurrent 38, 40 prerequisites 182
generic JVM arguments 173 start 186
getSharedClassHelperFactory() 63 submit PMR 187
group PTF 12, 132 troubleshoot 189
groupAccess 58, 61, 67 uninstall 186
update 184
IBM Technology for Java 76
H IBM Technology for Java VM
heap 6, 36, 41 installing 20
compaction 36, 39, 43, 46–47 verifying 22
contraction 42–44 IBM Technology for JVM 35, 42, 49–50, 73
default maximum size 42 memory constraints 110
default minimum size 42 IBM Toolbox for Java 10, 74–75
determining appropriate size settings 42 IBM_JAVACOREDIR 88
dump 112 idlj 8
expansion 42–43 IFS (Integrated File System) 81
fragmentation 36, 39, 46–47 initial heap size 15
avoiding through gencon GC policy 40 Integrated File System (IFS) 44
free space 41 Integrated Language Environment (ILE) 15
maximum interpreter 5
free space 44 invocation API 74
heap size 36 ISA 181
size 42 iSeries Navigator 131
theoretical size under Classic JVM 37
theoretical size under IBM Technology for JVM
37 J
minimum J2EE 3
free space 43 J2ME 3
heap size 36 J2SE 3
size 42 JAR file 50, 61
occupancy 46 Java
scanning 44 bytecode 4
settings 138 class libraries 2
tuning heap size 41 Development Kit 74
unlimited maximum size under Classic JVM 42 EE 3
HPROF 71 heap 6, 111
ME 3
method 2, 99
I runtime environment 50
i5/OS 36–37, 42, 50, 52, 60, 76, 95, 99, 110, 193 SDK 3
PASE 20 SE 3
installation 20 SecurityManager 67
native methods 80 shared class libraries 97
required PTFs 21 Software Development Kit 3
IBM Accelerator for System i5 128 virtual machine 4
IBM Developer Kit and Runtime Environment Java Native Interface (JNI) 73
Java 2 Technology Edition Java Platform 3
Version 5.0 Diagnostics Guide 193 Enterprise Edition 3
IBM Developer Kit for Java 12 Micro Edition 3
IBM developerWorks 47, 193 Standard Edition 3

204 IBM Technology for Java Virtual Machine in IBM i5/OS

java.hprof.txt 71 using 114
java.lang.ClassLoader 51 memory
java.net.URLClassLoader 51, 53–54, 63 allocation 36, 41
java.policy file 67 in gencon GC policy 40
JAVA_HOME 22 footprint 52
setting 23 of JVM 37
javac 8 freeing 36
javadoc 8 garbage collector 36
Javadump 92, 98 management 36
creating 93 memory allocator 36
files Memory Dump Diagnostic for Java 185
capturing 93 memory leak 111
importing into ThreadAnalyzer tool 95 minimum heap 31
JAVAGCINFO 30 mixed mode interpretation (MMI) environment 12
jdb 8 modification context 57, 61, 68, 71
JDBC 13
JDBC driver 130
jextract 192 N
JIT 6 native application 74
compiler 91 native heap 111
disabling 89 network 128
heap 7 noclassgc 133
recompilation threshold 7 nursery 40, 46
threshold 7 nursery space 40
JNI 6, 73–74
JNI (Java Native Interface) 6 O
job 75 object references 36–37
log 81 contraction in size of 37
JSR-163 194 object tenure age 40
just-in-time compiler 6 object-oriented language 2
JVAPGM objects 52 operating system security 67
JVM 74, 93 optavgpause 38–40
crash 110, 192 optimize for pause time 38–39
causes 87 optimize for throughput 38
hang optthruput 38–39
troubleshooting memory constraints 110 os400j9.systemlaunch.properties 176
launcher 55 OutOfMemory 112
overview of a hang 91 OutOfMemoryException 36
startup time 49
Tool Interface 6
troubleshooting a hang 92 P
JVMPI 142 paging 110
JVMTI 6, 68, 71, 142, 194 PASE 76
Performance Collection Services 128
Performance Explorer 15
K Performance Monitoring Infrastructure 174
kill -QUIT 93, 99 platform independent 4
PMAT 185
L PMI 174
L3 cache 128 portability 75
live objects 38, 40–41 prerequisites 20

M Q
machine interface 10 QEJBSVR user profile 67
maximum heap 31 QPFRADJ 130
MDD4J 185 Qshell 11
MDD4J tool 193 creating a Javadump in 93
launching 115 QWAS61 subsystem 99
memory requirements 114 QZDASOINIT 130

Index 205
R subpooling 38, 41, 48
RAMClass 53 subsystem 129
RAS 6 survivor space 40
Redbooks Web site 201 syntax 2
contact us x system API reference 76
Reliability, Availability, and Serviceability 6 system classloader 50–51, 54
requirements for i5/OS 168 system dump
response time 47 creating non-destructively 192
effect of garbage collection on 139 System i 41
ROMClass 53, 57, 63–64, 70–71 performance management Web site 129
RPG 74–75 System Licensed Internal Code 80
run() method 98 system.gc() 45
RUNJVA 12 SystemDefault.properties 25
running WebSphere Application Server
with new JVM 33 T
TCP/IP 13
S tenured space 40
scavenge 40, 46 teraspace storage model 76
SDK 3 thread
semantic 2 dump 98
Servlet 100 resource contention 93
Servlet handler threads 100 safety 76
shared class ThreadAnalyzer 95, 185
cache 50, 53 tool 92, 95, 193
metadata 53 analyzing Javadump using 94
permissions 67 monitor analysis 96
shared classes 49 summary 96
cache size 53 thread analysis 97
connecting to a class cache 53 threads 100, 103
deleting the class cache 58 throughput 47
deployment 54 effect of garbage collection on 139
dynamic cache update 59 effect of optthruput GC policy on 38
helper API 53–54, 63 Tivoli Performance Viewer 15
example 65 TMPDIR 88
Javadoc 66 Trade 6 performance benchmark sample 44, 133
history 52 transaction-based applications 40
listAllCaches suboption 55 troubleshooting tips 179
metadata 58, 62
overview 52 U
printAllStats suboption 57, 69 uninstalling 22
printStats suboption 56, 62 user-defined classloader 51, 68
recommendations 60 user-defined classloaders 53
security considerations 67 using the JAVAGCINFO 30
utility options 55
viewing class cache contents 57
viewing summary cache statistics 56 V
shared memory 52, 67 -verbose
shared semaphore 67 gc 44
SharedClassHelper 63 verbose garbage collection 132
SharedClassHelperFactory 63 verbose GC output 42
SharedClassTokenHelper 64 virtual machine 4
SharedClassURLClasspathHelper 64
SharedClassURLHelper 64
SLIC 80 W
source code 103 weak generational hypothesis 40
stale classes 59, 64 Web application 99
Stop The World 36 WebSphere Administrative Console 173
storeSharedClass() 65 WebSphere Application Server 12, 44, 51, 62, 67, 92, 98
subpool 41 default heap size 31
hang detection option 91

206 IBM Technology for Java Virtual Machine in IBM i5/OS

 shared classes usage in 54
troubleshooting hangs 91
using IBM Technology for JVM 27
WebSphere Development Studio Client for iSeries 17,
97–98
workload 43
estimator 128
WRKACTJOB 91
display job log using 93
WRKSYSSTS 111, 129

X
-Xdisableexplicitgc 45
-Xdump
heap 113
-Xgcpolicy 47
gencon 38
optavgpause 38
optthruput 38
subpool 38
-Xmaxf 44
-Xminf 43
XML 44
-Xms 42
-Xmx 42–43
-Xnoclassgc 133
XPCML 75
-Xscmx 59, 62
-Xshareclasses 54–56, 67, 69
-Xverbosegclog 44

Index 207
208 IBM Technology for Java Virtual Machine in IBM i5/OS
IBM Technology for Java Virtual
Machine in IBM i5/OS
IBM Technology for Java Virtual
Machine in IBM i5/OS
IBM Technology for Java Virtual Machine in IBM i5/OS
IBM Technology for Java Virtual Machine in IBM i5/OS
(0.2”spine)
0.17”<->0.473”
90<->249 pages
IBM Technology for Java Virtual
Machine in IBM i5/OS
IBM Technology for Java Virtual
Machine in IBM i5/OS
 Back cover ®

IBM Technology for

Java Virtual Machine in
IBM i5/OS
Most information This IBM Redbook gives a broad understanding of a new 32-bit Java
Virtual Machine (JVM) in IBM i5/OS. With the arrival of this new JVM, INTERNATIONAL
about 32-bit JVM
IBM System i platform now comfortably supports Java and WebSphere TECHNICAL
applies to all IBM
applications on a wide array of different server models: from entry size SUPPORT
server platforms boxes to the huge enterprise systems. ORGANIZATION
This book provides in-depth information about setting Java and IBM
Comprehensive book
WebSphere environments with new 32-bit JVM, tuning its
on new 32-bit JVM in performance, and monitoring or troubleshooting its runtime with the
i5/OS new set of tools.
BUILDING TECHNICAL
Information in this IBM Redbook helps system architects, Java INFORMATION BASED ON
Run WebSphere application developers, and system administrators in their work with PRACTICAL EXPERIENCE
Application Server in 32-bit JVM in i5/OS.
32-bit JVM Important: Despite the fact that this book targets i5/OS IBM Redbooks are developed
implementation, most information in this book applies to all IBM server by the IBM International
platforms, where the new 32-bit JVM is supported. Technical Support
Organization. Experts from
IBM, Customers and Partners
from around the world create
timely technical information
based on realistic scenarios.
Specific recommendations
are provided to help you
implement IT solutions more
effectively in your
environment.

For more information:

ibm.com/redbooks

SG24-7353-00 ISBN 0738486051