Mystery back to back Full GC

Question

Recently we've often seen 2 back to back GCs on our servers, and it seems both are stop the world GCs. The first GC happens toward the end of a young gen collection, the second happens right after and doesn’t seem to collect much memory. I cannot explain what's causing the first one, could it be overflow of survivor space during young gen collection causing full gc? If so, our old gen doesn't seem to be that full. If it is fragmentation issues, then i would be expecting to see some kind of promotion failure error messages. What puzzles me more is the second gc.
If anyone has any similar experiences or suggestions, I would really appreciate.
We are running on JDK 1.6.0_16 on Solaris. Here is the gc log for the two back to back Full GC:
37893.302: [GC 37893.302: [ParNew
Desired survivor size 153387008 bytes, new threshold 7 (max 7)
- age   1:   23471752 bytes,   23471752 total
- age   2:    9676008 bytes,   33147760 total
- age   3:   20601192 bytes,   53748952 total
: 745335K->166386K(1797568K), 0.0768924 secs]37893.379: [CMS: 4973695K->3666928K(8388608K), 5.1921440 secs] 5657754K->3666928K(10186176K), [CMS Perm : 53340K->53304K(86016K)], 5.2693834 secs] [Times: user=5.57 sys=0.04, real=5.27 secs]
37898.571: [Full GC 37898.571: [CMS: 3666928K->3666708K(8388608K), 4.3563034 secs] 3666928K->3666708K(10186176K), [CMS Perm : 53304K->53096K(86016K)], 4.3565658 secs] [Times: user=4.29 sys=0.02, real=4.36 secs]
Our JVM uses the following parameters:
-Xms10g
-Xmx10g
-XX:NewSize=2048m
-XX:MaxNewSize=2048m
-XX:+UseConcMarkSweepGC
-XX:+UseParNewGC
-XX:CMSMarkStackSize=64m
-XX:CMSInitiatingOccupancyFraction=80
-XX:+UseCMSInitiatingOccupancyOnly
-XX:SurvivorRatio=5
-XX:MaxTenuringThreshold=7
-XX:+DisableExplicitGC
-XX:+PrintTenuringDistribution
-XX:+PrintGCDetails
-XX:+PrintGCTimeStamps
-Xloggc:logs/gc.log

Answer 1

I'm also having this problem for full screen games. Hope they fix this soon, it's very aggravating.