On Tue, Jul 20, 2010 at 1:28 PM, Peter Schuller <peter.schul...@infidyne.com> wrote: >> Attaching Jconsole shows that there is a growth of memory and weird >> spikes. Unfortunately I did not take a screen shot of the growth of >> the spike over time. I'll do that when it occurs again. > > Note that expected behavior for CMS is to have lots of small ups and > downs as a result of young generation GC:s, and a longer period cycle > of larger ups and downs that correspond to CMS kicking in with a > concurrent mark/sweep phase. > > To get a reasonable estimate of the actual live set, you would > normally look at the free memory at the end of one of the larger dips > after a CMS mark/sweep phase.
You'll also see memory sawtoothing for memtable growth+flush cycles. -ryan