CMS collection keep working during holiday

[kirk]

Hi Ken,

Can I suggest that you back off on some of the switch settings. I think 
that GC might work better if ergonomics where allowed to adjust some of 
the sizes. This may prevent the horrible GC you are experiencing. Is 
suspect that an attempt at compaction in this condition (which the 
switch settings would exacerbate) would be devastating. I see 
SurvivorRatio=4 and then TargetSurvivorRatio=90. These completely 
conflicting settings. I don't know which one would win but I would guess 
SurvivorRatio=4. Some switch settings are fixing values where as other 
switch setting work as suggestions for ergonomics. Fixing values limit 
what ergonomics can do for you and ergonomics are designed to adjust 
configurations to help you meet your pause time and GC throughput goals. 
Parallel in new is default for 1.5+. For a first run I'd use....

-Xmx4 -XX:+UseConcMarkSweepGC + diagnostics + security settings

Regards,
Kirk

 


Ken-- at newsgroupstats.hk wrote:
> Thanks for your quick response. The application started at 1/10 was crash
> this morning. I lost the gc log file as I quickly restart the application to
> avoid service interruption to our clients.
>
> >From the hotspot error file(attached), it show that it crash when old
> generation 99% full. Anyway, it's another problem.
>
> http://www.nabble.com/file/p19841670/hs_err_pid2476.log hs_err_pid2476.log 
>
> I have another application with similar JVM options set and have exactly
> behavior:
>
> JVM Options:
>
> -Djava.security.policy=./java.policy -XX:+PrintGCDetails
> -XX:+PrintGCTaskTimeStamps -XX:+PrintHeapAtGC -Xloggc:hkexquoteserver_gc.log
> -Xms4G -Xmx4G -XX:NewSize=1G -XX:MaxNewSize=1G
> -XX:+PrintTenuringDistribution -XX:+DisableExplicitGC
> -XX:+UseConcMarkSweepGC -XX:+UseParNewGC -XX:SurvivorRatio=4
> -XX:CMSInitiatingOccupancyFraction=70 -XX:TargetSurvivorRatio=90
> -Dcom.sun.management.jmxremote.port=18003
> -Dcom.sun.management.jmxremote.authenticate=false
> -Dcom.sun.management.jmxremote.ssl=false 
>
> GC Logs:
> http://www.nabble.com/file/p19841670/gc.zip gc.zip 
>
> JConsole:
> http://www.nabble.com/file/p19841670/jconsole.jpg 
>
> http://www.nabble.com/file/p19841670/jconsole2.jpg 
>
>
> - long lasting CMSs from 426789 to 504376
> - CMS cannot free space in old generation within this period. What I
> expected is to free old-gen memory to around 500M full at 426789.
> - concurrent mode failure at 504379
>
> Thanks and Best Regards,
> Ken
>
>
>
>
>
>
>