1、dump命令行：
2、dump时机：CMS GC结束后，这时heap最小，Full GC也最快
2012-05-02T22:08:11.226+0800: 470959.079: [GC [1 CMS-initial-mark: 103404048K(121634816K)] 103797845K(125409728K), 0.2184710 secs] [Times: user=0.22 sys=0.00, real=0.22 secs]
2012-05-02T22:08:11.445+0800: 470959.298: [CMS-concurrent-mark-start]
...
2012-05-02T22:09:03.533+0800: 471011.386: [CMS-concurrent-mark: 51.042/52.088 secs] [Times: user=296.93 sys=5.87, real=52.08 secs]
2012-05-02T22:09:03.533+0800: 471011.386: [CMS-concurrent-preclean-start]
2012-05-02T22:09:04.432+0800: 471012.285: [CMS-concurrent-preclean: 0.855/0.899 secs] [Times: user=2.03 sys=0.05, real=0.90 secs]
2012-05-02T22:09:04.432+0800: 471012.285: [CMS-concurrent-abortable-preclean-start]
CMS: abort preclean due to time 2012-05-02T22:09:06.120+0800: 471013.973: [CMS-concurrent-abortable-preclean: 1.686/1.688 secs] [Times: user=3.69 sys=0.10, real=1.69 secs]
2012-05-02T22:09:06.137+0800: 471013.990: [GC[YG occupancy: 2860887 K (3774912 K)]471013.990: [Rescan (parallel) , 2.3901460 secs]471016.381: [weak refs processing, 0.0181930 secs]471016.399: [class unloading, 0.0240830 secs]471016.423: [scrub symbol & string
tables, 0.0026840 secs] [1 CMS-remark: 103739059K(121634816K)] 106599946K(125409728K), 2.4389880 secs] [Times: user=27.60 sys=0.13, real=2.44 secs]
2012-05-02T22:09:08.577+0800: 471016.430: [CMS-concurrent-sweep-start]
...
2012-05-02T22:10:49.798+0800: 471117.652: [CMS-concurrent-sweep: 98.440/101.222 secs] [Times: user=297.82 sys=10.19, real=101.21 secs]
2012-05-02T22:10:49.798+0800: 471117.652: [CMS-concurrent-reset-start]
2012-05-02T22:10:50.150+0800: 471118.004: [CMS-concurrent-reset: 0.352/0.352 secs] [Times: user=0.99 sys=0.03, real=0.35 secs]
3、dump内存会触发Full GC这时很悲剧的事情
2012-05-02T22:12:18.273+0800: 471206.127: [Full GC 471206.127: [CMS: 67912257K->65543780K(121634816K), 260.5313120 secs] 68920123K->65543780K(125409728K), [CMS Perm : 24838K->24823K(41476K)], 260.5314820 secs] [Times: user=259.15 sys=0.20, real=260.49 secs]
这次Full GC花了约5分钟。
4、Full GC结束后，立即开始dump，这时才能看到heap.bin文件在变大
从GC日志时间可以看出，持续了大约17分钟。dump的heap文件约65G，这样平均dump速度为：
70224599171MB(heap文件size)/1048s（dump时间）= 63.9MB/s
2012-05-02T22:34:06.579+0800: 472514.433: [GC 472514.433: [ParNew: 3355519K->370901K(3774912K), 0.2473880 secs] 68899300K->65914682K(125409728K), 0.2475830 secs] [Times: user=2.95 sys=0.02, real=0.24 secs]
5、这次dump内存工作总共花了22分钟，期间JobTracker完全不能工作，影响还是挺大的。
6、下一步期望能在Jobtracker上增加一个工具，dump出其主要内存数据。这样就大大减少了对用户的影响。
原文地址：http://blog.csdn.net/liangliyin/article/details/7529739