Java Capacity Planning

1
Java Capacity Planning Performance Measurements

• Dr. Carl J. De Pasquale
• carl_depasquale_at_adp.com

1
2
Java Capacity Planning Performance Measurements

• Agenda
• Capacity vs. Performance
• Java capacity planning
• Java black box
• Tools
• Java application
• Tools
• Performance Measurements
• Coding standards
• Byte code instrumentation
• Method performance measurements using management
  library
• Method performance measurements using ARM
  management library
• Tools
• Conclusions

3
Java Capacity Planning Performance Measurements

• Hardware Does Not Always Solve Capacity Problems
• Capacity of a given hardware configuration is
  subject to the application constraints
• Capacity f(Hardware, Application Performance)
• Assume a given hardware configuration
• is 70 utilized,
• Provides a 3 second application response,
• When the arrival rate is two hundred transactions
  per second.
• Management upgrades the hardware configuration
  with one that is twice as fast
• What could be the affect on the response time?

3
4
Java Capacity Planning Performance Measurements

• What could be the affect on the response time?
• Response Time Could Improve
• If there isnt an application CPU bottleneck
• If there isnt an application threading
  bottleneck
• If there isnt a network or I/O bottleneck
• Response Time Could Degrade
• Work arrives faster
• Saturates non CPU peripherals
• CPU experiences I/O Waits
• If there is an application threading bottleneck
• If SQL is a poorly written
• More physical I/O executed than required
• table scans/small reads
• If there is a Network Bottleneck
• Small packet size

4
5
Java Capacity Planning Performance Measurements

• Black box Java Capacity Planning
• The Java Process
• CPU
• Disk
• Network
• Memory
• No application visibility
• Many java classes running in one JVM
• Application Capacity Planning
• Process java
• Application Class/Method Data
• CPU time by method
• Elapsed time by method
• Method invocation count
• Memory usage heap
• Workflow trace
• SQL timing

5
6
Java Capacity Planning Performance Measurements

• Capacity Planning Black Box (java) - No
  Application Visibility
• Collect process data
• Identify busy period
• Analyze data
• Review reports to identify hot spots
• Computer CPU utilization
• Computer summary
• Disk IO summary
• Network summary
• Memory summary
• Process summary Java will usually be the top
  resource consumer

6
7
Java Capacity Planning Performance Measurements

• Capacity Planning Black Box (java) - No
  Application Visibility
• Identify and eliminate all bottlenecks
  (analytically)
• Apply and monitor changes to physical
  infrastructure
• Once bottlenecks are eliminated
• Recollect performance data
• Use Analytical modeling to help determine
• Capacity f(Hardware, Application Performance)
• Tools
• Opnet IT GURU
• Hyperformix Capacity Manager
• Team Quest
• Performance Assurance

7
8
Java Capacity Planning Performance Measurements

• Application Capacity Planning (java) -
  Application Visibility
• Coding Standards
• Application Specific Log
• Non standard format
• Each application log must be parsed by capacity
  planner
• May or may not contain necessary data
• Better than nothing
• Tools
• Byte Code Engineering Library (BCEL)
• Java ClassLoader

8
9
Java Capacity Planning Performance Measurements

• Sample Code
• public final class test
• test ()
• public void runtest ()
• for ( int i 0 i lt 9999999 i )
• public static void main(String argv)
• try
• for (int i 0 i lt 1 i)
• test t1 new test()
• t1.runtest()
• for (int j 0 j lt 2 j)
• new oohrah ( argv0, argv1 )
• catch ( Exception e )
• e.printStackTrace()

• Sample Code
• public class oohrah
• public oohrah( String a, String b )
• spin()
• public static void spin()
• int i
• new hello()
• for (i0 i lt 500000000 i)
• public class hello
• public hello( )
• printHello()
• public static void printHello()

9
10
Java Capacity Planning Performance Measurements

• Dynamic Byte Code Instrumentation - JDK 1.5

10
11
Java Capacity Planning Performance Measurements

• Dynamic Byte Code Instrumentation - JDK 1.5
• Start Class Nametest,Method Name main Elapsed
  1180537593616 CPU 515625000
• Start Class Nametest,Method Name runtest Elapsed
  1180537593632 CPU 515625000
• End Class Nametest,Method Name runtest Elapsed
  1180537593663 CPU 546875000
• ClassNametest runtestCountgt 1
• ClassNametest runtestCpugt 31250000
• ClassNametest mainCountgt 1
• ClassNametest mainCpugt 0
• Start Class Nameoohrah,Method Name spin Elapsed
  1180537593679 CPU 562500000
• Start Class Namehello,Method Name printHello
  Elapsed 1180537593679 CPU 562500000
• HelloWorld (from hello)
• End Class Namehello,Method Name printHello
  Elapsed 1180537593679 CPU 562500000
• ClassNamehello printHelloCountgt 1
• ClassNamehello printHelloCpugt 0
• End Class Nameoohrah,Method Name spin Elapsed
  1180537595053 CPU 1890625000
• ClassNameoohrah spinCountgt 1
• ClassNameoohrah spinCpugt 1328125000
• Start Class Nameoohrah,Method Name spin Elapsed
  1180537595053 CPU 1890625000
• Start Class Namehello,Method Name printHello
  Elapsed 1180537595053 CPU 1890625000

11
12
Java Capacity Planning Performance Measurements

• Application Capacity Planning (java) -
  Application Visibility
• ARM Instrument
• Elapsed time
• Invocation counts
• Correlation
• ArmTransactionsWithMetrics
• CPU time
• Tools
• Byte Code Engineering Library (BCEL)
• Java ClassLoader

12
13
Java Capacity Planning Performance Measurements

• Dynamic Byte Code Instrumentation - ARM

13
14
Java Capacity Planning Performance Measurements

• Dynamic Byte Code Instrumentation ARM
• arm_start_transaction()
• time030934 06/13/07
• transaction id02000000 00000000 00000000
  00000000
• parent correlator none
• flagsNONE
• application nameApplicationDefinition test
• transaction nameTransactionDefinition
  test.main
• transaction instance handle1

14
15
Java Capacity Planning Performance Measurements

• Dynamic Byte Code Instrumentation ARM
• arm_start_transaction()
• time030934 06/13/07
• transaction id04000000 00000000 00000000
  00000000
• parent correlator
• length44
• format126
• flags0
• app nameApplicationDefinition test
• app handle1
• tran nameTransactionDefinition test.main
• tran instance id1
• flagsNONE
• application nameApplicationDefinition test
• transaction nameTransactionDefinition
  test.runtest
• transaction instance handle2
• arm_stop_transaction()
• time030934 06/13/07

15
16
Java Capacity Planning Performance Measurements

• Dynamic Byte Code Instrumentation ARM
• arm_start_transaction()
• time030934 06/13/07
• transaction id06000000 00000000 00000000
  00000000
• parent correlator
• length44
• format126
• flags0
• app nameApplicationDefinition test
• app handle1
• tran nameTransactionDefinition test.main
• tran instance id1
• flagsNONE
• application nameApplicationDefinition oohrah
• transaction nameTransactionDefinition
  oohrah.spin
• transaction instance handle3

16
17
Java Capacity Planning Performance Measurements

• Dynamic Byte Code Instrumentation ARM
• arm_start_transaction()
• time030934 06/13/07
• transaction id08000000 00000000 00000000
  00000000
• parent correlator
• length44
• format126
• flags0
• app nameApplicationDefinition oohrah
• app handle3
• tran nameTransactionDefinition oohrah.spin
• tran instance id3
• flagsNONE
• application nameApplicationDefinition hello
• transaction nameTransactionDefinition
  hello.printHello
• transaction instance handle4

17
18
Java Capacity Planning Performance Measurements

• Dynamic Byte Code Instrumentation ARM
• arm_stop_transaction()
• time030934 06/13/07
• transaction instance handle4
• transaction statusGOOD
• flagsNONE
• application nameApplicationDefinition hello
• transaction nameTransactionDefinition
  hello.printHello
• arm_stop_transaction()
• time030935 06/13/07
• transaction instance handle3
• transaction statusGOOD
• flagsNONE
• application nameApplicationDefinition oohrah
• transaction nameTransactionDefinition
  oohrah.spin

18
19
Java Capacity Planning Performance Measurements

• Dynamic Byte Code Instrumentation ARM
• arm_start_transaction()
• time030935 06/13/07
• transaction id0a000000 00000000 00000000
  00000000
• parent correlator
• length44
• format126
• flags0
• app nameApplicationDefinition test
• app handle1
• tran nameTransactionDefinition test.main
• tran instance id1
• flagsNONE
• application nameApplicationDefinition oohrah
• transaction nameTransactionDefinition
  oohrah.spin
• transaction instance handle5

19
20
Java Capacity Planning Performance Measurements

• Dynamic Byte Code Instrumentation ARM
• arm_start_transaction()
• time030935 06/13/07
• transaction id0c000000 00000000 00000000
  00000000
• parent correlator
• length44
• format126
• flags0
• app nameApplicationDefinition oohrah
• app handle5
• tran nameTransactionDefinition oohrah.spin
• tran instance id5
• flagsNONE
• application nameApplicationDefinition hello
• transaction nameTransactionDefinition
  hello.printHello
• transaction instance handle6

20
21
Java Capacity Planning Performance Measurements

• Dynamic Byte Code Instrumentation ARM
• arm_stop_transaction()
• time030935 06/13/07
• transaction instance handle6
• transaction statusGOOD
• flagsNONE
• application nameApplicationDefinition hello
• transaction nameTransactionDefinition
  hello.printHello
• arm_stop_transaction()
• time030935 06/13/07
• transaction instance handle5
• transaction statusGOOD
• flagsNONE
• application nameApplicationDefinition oohrah
• transaction nameTransactionDefinition
  oohrah.spin
• arm_stop_transaction()
• time030935 06/13/07

21
22
Java Capacity Planning Performance Measurements

• Analyzing the Raw Data

22
23
Java Capacity Planning Performance Measurements

• Operational Analysis Method Raw Data Process

23
24
Java Capacity Planning Performance Measurements

• Operational Analysis Overloading
• Process ? Method
• Method ? Process
• Valid for
• Black box, and application capacity planning
• Denning, and J. Buzen, The operational analysis
  of queuing network models. Computer Surveys,
  10(3), 225 - 261 (1978)

24
25
Java Capacity Planning Performance Measurements

• Third Party Performance Analysis Tools
• Unable to implement custom application logging
• Unable to implement Dynamic Byte Code
  Instrumentation
• Third party alternatives
• Mercury
• Opnet Panorama/ACE
• Quest Perform Assure
• Wily

25
26
Java Capacity Planning Performance Measurements

• Conclusion
• By adding a bit of sophistication during
  application development, performance testing, or
  production deployment the often overlooked
  non-functional requirement of application
  performance requirement can be addressed

26
27
Java Capacity Planning Performance Measurements

• Thank You for Attending Have a nice rest of the
  day

Questions
Comments
Discussion
27