garbage collector locks up

Performance Tuning
#1

to make it short:
tomcat, arguments:
-Xmx513m
-Xms512m
-server
-XX:+UseParNewGC
-XX:+CMSParallelRemarkEnabled
-XX:+UseConcMarkSweepGC
-XX:+PrintGCDetails
-verbosegc
-XX:+CMSPermGenSweepingEnabled
-XX:+CMSClassUnloadingEnabled

runs fine for a while, then prints:

[GC [1 CMS-initial-mark: 261312K(516096K)] 261420K(524224K), 0.0008936 secs]
[CMS-concurrent-mark: 0.673/0.673 secs]
[CMS-concurrent-preclean: 0.000/0.000 secs]
[GC[YG occupancy: 107 K (8128 K)][Rescan (parallel) , 0.0034411 secs][weak refs processing, 0.0000064 secs][class unloading, 0.0214873 secs][scrub symbol & string tables, 0.0053496 secs] [1 CMS-remark: 261312K(516096K)] 261420K(524224K), 0.0353675 secs]
[CMS-concurrent-sweep: 0.186/0.186 secs]
[CMS-concurrent-reset: 0.009/0.009 secs]

over and over again, consuming 50% of my hyperthreading p4.

sometimes i get:
[Full GC [CMS (concurrent mode failure): 261308K->261317K(516096K), 1.3826536 secs] 261493K->261317K(524224K), [CMS Perm : 65535K->65534K(65536K)], 1.3828157 secs]
[GC [1 CMS-initial-mark: 261317K(516096K)] 261317K(524224K), 0.0008784 secs]
[Full GC [CMS[CMS-concurrent-mark: 0.621/0.623 secs]
(concurrent mode failure): 261317K->261315K(516096K), 2.0470819 secs] 261502K->261315K(524224K), [CMS Perm : 65535K->65535K(65536K)], 2.0472546 secs]
[Full GC [CMS (concurrent mode failure): 261315K->261315K(516096K), 1.1678882 secs] 261368K->261315K(524224K), [CMS Perm : 65535K->65535K(65536K)], 1.1680489 secs]
[GC [1 CMS-initial-mark: 261315K(516096K)] 261315K(524224K), 0.0011766 secs]
[Full GC [CMS[CMS-concurrent-mark: 0.686/0.686 secs]
(concurrent mode failure): 261315K->261314K(516096K), 1.9247622 secs] 261368K->261314K(524224K), [CMS Perm : 65535K->65535K(65536K)], 1.9249726 secs]
[Full GC [CMS (concurrent mode failure): 261314K->261314K(516096K), 1.2463526 secs] 261314K->261314K(524224K), [CMS Perm : 65535K->65535K(65536K)], 1.2464900 secs]
[GC [1 CMS-initial-mark: 261314K(516096K)] 261314K(524224K), 0.0008861 secs]

again and again. whats wrong?

#2

Hmm. all the -XX flags are “use at your own risk”. My guess is you have a combination of them that is causing the VM to do someting bad.

But exactly what that combination is, I wouldn’t know. Maybe one of the VM guys like Ken or Azim will take a loo kat your post and have more to say.

Otherwise Id just start eliminating -XX flags and see where the problem goes away.

#3

You’ll probably get better answers if you post on the JVM forum on forum.java.sun.com.

It looks to me at least in the second case that you’ve loaded a ton of classes. You might want to try increasing the perm gen size with -XX:MaxPermSize=128m or similar.

#4

is 3000+ a ton?

#5

3000+ classes?

#6

yes, >3000 different classses
actually, it’s > 3000 files, so it’s even more.

#7

It’s hard to tell. I think NetBeans increases the perm gen size on the command line and they load some 5000+ classes.

#8

I thought the perm gen size automatically resized as necessary anyway?

Cas :slight_smile:

#9

Only up to a certain maximum, just like the rest of the heap. The default MaxPermSize is 64 MB.