Necessary waste collection help

Question

I am looking for some guidance trying to diagnose a problem with the GC.

We are testing Solaris using WebSphere Portal, and my current environment has an overhead of garbage collection of 7% (this was calculated using a detailed GC and running the log via PMAT). I have to compare this value with another environment that works at an average level of 4.5%. Envs are on the same version of WebSphere Portal, the same JVM sizes, parameters, custom varibles, etc. My JVMs had 20 more distribution failures, 2 more full GCs, average pause time per second per GC than their environment, during a 1 hour performance testing period.

Can you give me any advice on what might cause this problem with all the same configuration values ​​and the same test for 1 hour of performance? Or anything else for a review?

thanks

Answer 1

The first thing I would like to check out is that this process is ever inserted into swap / "virtual" memory. This will have a significant impact on your GC overhead.

With identical Java configurations and loading, etc. the most obvious difference will be the amount of physical memory on each machine. Assuming they are the same, see what other processes / services are running on each machine - unless they are new installations of the same OS, I expect there will be some (potentially significant) differences.

Send the output from top and iostat (or your preferred tools) and we will see if there is anything obvious.

Answer 2

If you work with different operating systems, it is possible that they work differently with the same settings. Optimum GC settings and threadpool sizes may vary. I remember that at least in the case of Solaris vs RedHat, the optimal concurrency settings were clearly different.