最近我们在一些大型Silicon Graphics服务器上对JRockit进行了测试。下面是其中一次测试的部分日志：

 ...
 [memory ] GC strategy: parallel
 [memory ] using large pages - maximal oldspace size commited at startup
 [memory ] heap size: 468992000K, maximal heap size: 469237760K
 ...

　　 差不多有450 GB的堆。是的——Java堆！我家里计算机的硬盘都没有这么大。

　　 好消息是它运行正常。甚至其GC暂停时间也不是太长：一次full GC大约耗时一分钟。

评论

• 机器上有多少CPU（对并行gc）？

  　　 能否贴出一些GC日志片断和你们所得到的虚拟机最优设置选项？

　　发表人：moazam，2006年5月4日，02:02 PM

• 我认为是这个128P服务器。关于如何调优大型机器的JVM，我会单独写一篇文章。

　　下面是GC日志：

[memdbg ] old collection 24 started
[memdbg ] OC reasons: Large obj: 1 (4816 bytes), TLA:0-Webinar:0-Webinar:1-Webinar:[BEA0: 0
[compact] External compact ratio set to: 8
[compact] Internal compact ratio set to: 8
[memdbg ] Compacting 8 heap parts at index 67 (type 1) (exceptional 0)
[compact] OC 24: 8 parts (max 128), index 67. Type 1, exceptional 0
[compact] Area start: 0x8000003a87660000, end: 0x8000004184760000
[memdbg ] starting parallel marking phase
[memdbg ] Pointerset limit hit, compaction aborted.
[compact] Pointerset at index 30 full.
[gcpause] total mark time: 1037.667 ms
[memdbg ] ending marking phase
[memdbg ] starting parallel sweeping phase
[gcpause] total sweep time: 1051.062 ms
[memdbg ] ending sweeping phase
[compact] Compaction pause: 0.000 (target 1281.445), update ref pause: 0.000 (target 1281.445)
[compact] Updated 243 refs (limit:3809).
[compact] Summary: 24;67;8;0;0;0.000;1281.445;0.000;1281.445;243;3809
[gcpause] thread waited for memory 2148.917 ms
[gcpause] old collection phase 0 pause time: 2091.180000 ms, (start time: 2727.152 s)
[gcpause] (pause includes compaction: 0.000 ms (no compaction), update ref: 0.000 ms)
[memory ] 2727.152-2729.244: GC 468992000K->117688605K (468992000K), 2091.180 ms
[memdbg ] Page faults before GC: 0, page faults after GC: 0, pages in heap: 29312000
[finaliz] (OC) Pending finalizers 0->0
[memdbg ] old collection 25 started
[memdbg ] OC reasons: Large obj: 1 (4816 bytes), TLA:0-Webinar:0-Webinar:1-Webinar:[BEA0: 0
[compact] External compact ratio set to: 8
[compact] Internal compact ratio set to: 8
[memdbg ] Compacting 8 heap parts at index 91 (type 2) (exceptional 0)
[compact] OC 25: 8 parts (max 128), index 91. Type 2, exceptional 0
[compact] Area start: 0x8000004f7e960000, end: 0x800000567ba60000
[memdbg ] starting parallel marking phase
[gcpause] total mark time: 1210.143 ms
[memdbg ] ending marking phase
[memdbg ] starting parallel sweeping phase
[compact] Evacuation reached low watermark due to timeout.
[compact] Evacuation interrupted after 1441.387000 ms at 0x80000051824d9d80
[compact] Evacuation added 21365704736 bytes to freelist
[compact] Evacuation found 200 objects and moved 200 objects
[compact] Updated references in 844 nonmoved and 200 moved objects
[compact] Compaction overhead increased to: 34.172
[gcpause] total sweep time: 5128.479 ms
[memdbg ] ending sweeping phase
[compact] Average compact time ratio: 0.566053
[compact] Compaction pause: 2031.806 (target 1922.168), update ref pause: 2049.646 (target 1922.168)
[compact] Updated 543 refs (limit:3809).
[compact] Summary: 25;91;8;2;0;2031.806;1922.168;2049.646;1922.168;543;3809
[gcpause] thread waited for memory 6412.586 ms
[gcpause] old collection phase 0 pause time: 6376.230000 ms, (start time: 2777.638 s)
[gcpause] (pause includes compaction: 2031.806 ms (external), update ref: 2049.646 ms)
[memory ] 2777.638-2784.014: GC 468992000K->109506404K (468992000K), 6376.230 ms
[memdbg ] Page faults before GC: 0, page faults after GC: 0, pages in heap: 29312000
[finaliz] (OC) Pending finalizers 0->0

　　发表人：hstahl，2006年5月4日，11:39 PM

原文出处：http://dev2dev.bea.com/blog/hstahl/archive/2006/05/what_would_you.html