Locating Performance Regressions in Code

1
Locating Performance Regressions in Code

• Peter Morong
• Doc. Ing. Petr Tuma, Dr

2
introduction

• investigate performance changes in code
• design framework
• define methods tools
• test real projects

3
profiling

• performance influence
• profiler requirements
• measured software modules

4
methods description

• source code changes categorization
• direct scope / indirect scope
• Advanced difference filtering
• Comparing profiler results

5
visualization tool

• requirements
• prototype implementation
• OProfile reader, HTML generator

Input Parser
Main Processor
Output Generator
XML
XML
6
MONO

• open source .NET implementation
• internal profiler
• monocov Mono Code Coverage
• 'no other such tool for mono' Zoltan Varga

7
CORBA

• omniORB, TAO
• OProfile
• both methods successful
• Advanced difference filtering
• 46 changed / 54 profiled / 4 files (omniORB)?
• 131 changed / 314 profiled / 13 files (TAO)?

8
omniORB
9
visualization tool
10
visualization tool
11
JAVA

• JORAM, ProActive
• HPROF, JAMon, JRat, JMP, JIP
• CodeCover, JVMDICover, EMMA, Cobertura
• one method successful
• Advanced difference filtering
• 79 changed / 68 profiled / 17 files (JORAM)?

12
conclusion

• investigate performance changes
• design framework (create new methods)?
• test MONO impl.
• test CORBA impl.
• test Java projects
• Open issues other profilers/systems (VM),
  improve visualization tool

13
Thank you

• Any questions?