GC time doubles in 1.5 and doesn't happen in 1.4.2 with same heap settings

Question

I have an application which runs in 1.4.2 and 1.5.0_05 JVM. Below are my heap settings which is same in both the versions.
-Xms2048m -Xmx2048m -XX:+PrintGCDetails
-XX:NewSize=256m -XX:MaxNewSize=256m
-XX:PermSize=64m -XX:+UseParNewGC
-XX:ParallelGCThreads=4 -XX:+UseConcMarkSweepGC
-XX:+CMSPermGenSweepingEnabled
-XX:CMSInitiatingOccupancyFraction=80
-XX:+CMSClassUnloadingEnabled
-XX:+UseCMSInitiatingOccupancyOnly
-XX:+CMSParallelRemarkEnabled
-XX:MaxTenuringThreshold=0 -XX:SurvivorRatio=128
Between every GC there is an increase of 3MB in heap size in both cases. Occasionally there is an increase of 11 MB to 20 MB in heap size between GC's in both cases.
Below is a sample o/p from gc logs for 1.5 JVM. The bold portion represents a jump in GC time (approximately doubles) as the heap increases by 11 MB (marked in bold). During this increase in heap size the GC time doubles and never reverts. The timing stays there
250.839:*[GC*250.839:*[ParNew:*258176K->0K(260160K),*0.0721558*secs]*630891K->376276K(2095168K),*0.0724650*secs]
252.979:*[GC*252.980:*[ParNew:*258176K->0K(260160K),*0.0733005*secs]*634452K->379857K(2095168K),*0.0736051*secs]
255.304:*[GC*255.304:*[ParNew:*258175K->0K(260160K),*0.1298656*secs]*638033K->391535K(2095168K),*0.1302190*secs]
257.525:*[GC*257.525:*[ParNew:*258176K->0K(260160K),*0.1435846*secs]*649711K->395126K(2095168K),*0.1438980*secs]
259.790:*[GC*259.790:*[ParNew:*258176K->0K(260160K),*0.1428786*secs]*653302K->398703K(2095168K),*0.1431886*secs]
Below is the gc logs from 1.4.2 JVM for the same application with same heap settings. Eventhough there is an increase of 11 MB in heap size and the GC time doubles, the GC time reverts to the same value as before the heap size got increased.
222.174:*[GC*222.174:*[ParNew:*261760K->0K(261952K),*0.0529798*secs]*640319K->382076K(2096960K),*0.0531960*secs]
224.186:*[GC*224.186:*[ParNew:*257968K->0K(261952K),*0.0521684*secs]*640045K->385561K(2096960K),*0.0524359*secs]
226.237:*[GC*226.237:*[ParNew:*261760K->0K(261952K),*0.1865245*secs]*647321K->397045K(2096960K),*0.1867803*secs]
228.340:*[GC*228.340:*[ParNew:*261760K->0K(261952K),*0.0537954*secs]*658805K->400488K(2096960K),*0.0540503*secs]
230.324:*[GC*230.324:*[ParNew:*261760K->0K(261952K),*0.0524924*secs]*662248K->403890K(2096960K),*0.0527366*secs]
Are there any heap settings that I'm missing or are there any issues with 1.5 JVM ?
thanks

Answer 1

It would be helpful if we could see some context for the collections that have
doubled in durations (also for the equivalents ones with 1.4.2 that did not
double in duration). In particular, is there a concurrent collection in progress?
If you have more complete logs that you would like to provide, submit a bug
report as ramki_at_jdc suggests and attach the logs to the bug report.
Does your application have objects that are mostly of a particular size? Or
mostly objects of a few sizes?