Erratic(?) CMS behaviour after 5d

Bernd Eckenfels bernd-2012 at
Sat Sep 29 21:03:12 UTC 2012


> You're seeing concurrent mode failures (see attached graph)

The graph is cool, how do you create it. But I am not sure about the 
concurrent Mode failure theory, there is no such message in the log, the
old heap is nearly empty and no FullGC are happening.

> i'd recommend trying these flags:
> -XX:CMSInitiatingOccupancyFraction=60

Yes, currently it is much smaller by auto tuning (I guess I will use 10%
and shrink the heap. AFAIK it needs to be combined with

> It's also possible you are running out of old gen space,
> try increasing your old gen size as well.

Do you know if it is visible in jstat -gccause if Perm is the reason? According
to jstat -heap it was only 75% filled, but I also treat it as a likely

> > concurrent mark-sweep generation:
> >      capacity = 25769803776 (24576.0MB)
> >      used     = 1704634416 (1625.6660614013672MB)
> >      free     = 24065169360 (22950.333938598633MB)
> >      6.614852137863636% used
> > Perm Generation:
> >      capacity = 314572800 (300.0MB)
> >      used     = 238131048 (227.0994644165039MB)
> >      free     = 76441752 (72.9005355834961MB)
> >      75.69982147216797% used


PS: lets discuss that in -use list.

More information about the hotspot-gc-dev mailing list