Extremely long parallel rescan

[Y Srinivas Ramakrishna]

Can you run with -XX:PrintCMSStatistics=2 and send me the output?

I suspect a performance badness related to code for
manipulating overflow lists that we recently discovered
and which is being fixed under:

6786503 Overflow list performance can be improved
6787254 Work queue capacity can be increased substantially on some platforms

We can get you a fix verification binary
through your support channels once it is established
that work queue overflow is the issue here (and it
sounds very much like it is).

-- ramki

----- Original Message -----
From: Hi Guava <higuava at gmail.com>
Date: Thursday, January 15, 2009 9:27 am
Subject: Extremely long parallel rescan
To: hotspot-gc-use at openjdk.java.net


> We have been having problem with CMS collector for a long time. The rescan
> phase of CMS takes very very long time (20 minutes to several hours). 
> We
> tried searching on Internet, asking in Sun's developer forum, etc. No 
> one
> seems know what's going on.
> 
> The typical hardware has 4 to 8 cpu/cores with 20 GB to 64GB of 
> memory. It
> happens on both Windows and Linux version of 64-bit JVMs.
> The GC parameters that we usually use are: -XX:+UseConcMarkSweepGC
> -XX:+PrintGCDetails -XX:+PrintGCTimeStamps
> 
> The following is an example of the problem.
> Hardware: 2 dual-core CPUs (4 core in total) with 28 GB memory.
> JVM: Java HotSpot(TM) 64-Bit Server VM Version 1.5.0_17-b04 from Sun
> Microsystems Inc. on Windows XP Version 5.2 on amd64
> Settings: -Xms20000m -Xmx20000m -XX:NewSize=500m
> -XX:+CMSPermGenSweepingEnabled -XX:+DisableExplicitGC
> -XX:CMSMarkStackSize=2M -XX:CMSMarkStackSizeMax=8M -XX:+UseConcMarkSweepGC
> -XX:CMSInitiatingOccupancyFraction=30 -XX:+PrintGCDetails
> -XX:+PrintGCTimeStamps
> Note that we set many other GC parameters in this particular example.
> Usually we only use -XX:+UseConcMarkSweepGC.
> 
> 910.659: [GC [1 CMS-initial-mark: 6465020K(19968000K)] 6679872K(20479552K),
> 0.3206026 secs]
> 910.980: [CMS-concurrent-mark-start]
> 919.131: [GC 919.131: [ParNew: 511104K->0K(511552K), 0.6801310 secs]
> 6976124K->6657978K(20479552K), 0.6802201 secs]
> 933.413: [CMS-concurrent-mark: 21.739/22.433 secs]
> 933.413: [CMS-concurrent-preclean-start]
> 934.813: [CMS-concurrent-preclean: 1.348/1.400 secs]
> 934.813: [CMS-concurrent-abortable-preclean-start]
>  CMS: abort preclean due to time 935.929:
> [CMS-concurrent-abortable-preclean: 0.662/1.116 secs]
> 935.933: [GC[YG occupancy: 443765 K (511552 K)]935.933: [Rescan (parallel)
> , 3533.9840087 secs]4469.917: [weak refs processing, 0.0620489 secs] [1
> CMS-remark: 6657978K(19968000K)] 7101744K(20479552K), 3534.0462074 secs]
> 4469.980: [CMS-concurrent-sweep-start]
> _______________________________________________
> hotspot-gc-use mailing list
> hotspot-gc-use at openjdk.java.net
> http://mail.openjdk.java.net/mailman/listinfo/hotspot-gc-use
_______________________________________________
hotspot-gc-use mailing list
hotspot-gc-use at openjdk.java.net
http://mail.openjdk.java.net/mailman/listinfo/hotspot-gc-use