Memory Measurement Discrepancy

[Mike Hawk]

Why is it that the memory measurement one gets from verbose garbage collection is different from measurements for the jvm process?
We used third-party packages and UNIX system commands to measure memory for the jvm process, and they do correlate at all with the memory reported by the garbage collector.

[Ilja Preuss]

Originally posted by Mike Hawk:
Why is it that the memory measurement one gets from verbose garbage collection is different from measurements for the jvm process?

I would simply expect the JVM to need more memory than just the memory for the java object heap. There has to be memory for the stack, memory for the bytecode, memory for the native code, for the hotspot engine etc. All this memory has nothing to do with gc.

[Mike Hawk]

Thanks for your reply. I don't think I adequately stated the problem. We are running AIX 4.3.3. We run svmon to record the process's memory usage as seen by the operating system. When we compare the svmon output to the output from gc, there appears to be no correlation: Sometimes the gc reports the heap size is bigger than what the OS thinks the process sometimes smaller.
But we've just eyeballed the tabular reports. I think my next trick will be to run a statistical correlation of the svmon output with the gc output to see if there is any relationship.
Thank again,