JMP - Java Memory Profiler |
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
General About Installation Licence Changelog Screenshot Performance
Documentation
Download
Contacts |
Performance Depending on how you run jmp it takes between 1.5 and 20 times as long for the java program to complete. JVMPI is quite slow, enabling all events that jmp uses can make the program take up to 10 times as long. There is a big difference between the available jvms, for example the IBM jvm has previously done a lot more inlining even when jvmpi was enabled causing it to be much faster for profiling. The tests done here are artificial. You do not want to run with full profiling from the start. You do want to run with minimal profiling and when you have reached an interesting state turn on object tracing or method tracing. Note that the testing done here is only an indication on how fast (or slow) jmp is, your milage may vary. The standard test case for jmp is running the RabbIT proxy on localhost. And then time a wget for 31 standard url (jmp website) from the local webserver, like this: Timing table for this test under jmp/0.41 and RabbIT/2.0.29. Test machine is an AMD athlon 2.8 GHz. 512 MB RAM using Linux/2.6.4. Running debian/GNU linux unstable branch. The webserver is a standard apache/2.0.48
The test performes a number of operations:
Old results Testing on jmp/0.14 and rabbit/2.0.16 can be found |