Gencon causes erratic performance on SMP linux server

Question

-----Problems:
1. Erratic performance.
2. GC logs show only half of the GC pause times.
-----Symptoms:
Long pause(s) in application response time. GC log output shows GC collections lasting about half the duration of the application pause. Our testing resulted in pauses which lasted up to 90 seconds during which few or no transactions were processed by the workload.
-----Cause:
Gencon apparently does not work well on an MP Linux system
-----Workaround:
Use the parallel gc option on MP systems.
-----System Specs:
Red Hat 3.1 AS
Xeon MP (quad)
Standalone java application on JRockit.
Java(TM) 2 Runtime Environment, Standard Edition (build 1.4.2_04-b05)
BEA WebLogic JRockit(TM) 1.4.2_04 JVM (build ari-31788-20040616-1032-linux-ia32, Native Threads,

Answer 1

David,
Thank you for your response. I believe the pauses are GC related, though.
The application is multithreaded and can be run in multiple instances as well. This behavior effects one entire process. That is, all threads in one process stop producing for several seconds (up to 90 as I mentioned in my original post). Other instances of the application on the same server continue to function normally.
If the GC option is changed to parallel then this pause behavior does not occur. Changing the GC option was made in lab conditions and was the only change made. Reverting to gencon, of course, reliably reproduces the problem.
Additionally the timing of the longest GC pause in the GC log correlates with the timing of the transactional drop in the application's log. Again, as I mentioned in the first post, the time of the GC pause is half of the time of the unresponsive period in the application. Therefore, even if the application is causing half of the delay, a 45 second GC pause for a 1.5 GB heap is cause for review.
Thank you again for your attention to this matter,
Dan