"-XX:MaxTenuringThreshold=0" - Tenuring Objects Immediately

Java GC Tutorials - Herong's Tutorial Examples

∟Concurrent Mark-Sweep (CMS) Collector - "+XX:+UseConcMarkSweepGC"

∟"-XX:MaxTenuringThreshold=0" - Tenuring Objects Immediately

This section provides a tutorial on how to use '-XX:MaxTenuringThreshold=0' to force objects being promoted immediately to Tenured generation. This will result more Major GCs.

Earlier, we learned how to give a large Young generation to limit the GC in the Young generation using the ParNew collector only.

In this tutorial, let's try to use the "-XX:MaxTenuringThreshold=0" JVM option to perform Tenured generation GC and Young generation GC:

herong> java -XX:+UseConcMarkSweepGC \
   -Xms80m -Xmx80m -XX:NewRatio=3 -XX:SurvivorRatio=18 \
   -XX:MaxTenuringThreshold=0 \
   -Xlog:gc,gc+heap=debug,gc+age=trace \
   GarbageCollection2 > output.txt

herong> more output.txt

Using Concurrent Mark Sweep
Step/TotalMemory/FreeMemory/UsedMemory:
1   82837504   80656320   2181184
...
16   82837504   64172304   18665200
GC(0) Desired survivor size 524288 bytes, new threshold 0 (max threshold 0)
GC(0) Age table with threshold 0 (max threshold 0)
GC(0) ParNew: 18227K->0K(19456K)
GC(0) CMS: 0K->12104K(61440K)
GC(0) Pause Young (Allocation Failure) 17M->11M(79M) 10.493ms
   -- GC #0 - Young generation only
   -- All live objects were promoted directly
   -- from "eden" space to "tenured" generation
   -- because of -XX:MaxTenuringThreshold=0

...
33   82837504   52246896   30590608
GC(1) ParNew: 17769K->0K(19456K)
GC(1) CMS: 12104K->24392K(61440K)
GC(1) Pause Young (Allocation Failure) 29M->23M(79M) 6.775ms
   -- GC #1 - Young generation only
   -- More live objects were promoted directly to the Tenured generation

...
50   82837504   39662368   43175136
GC(2) ParNew: 17770K->0K(19456K)
GC(2) CMS: 24392K->36680K(61440K)
GC(2) Pause Young (Allocation Failure) 41M->35M(79M) 6.169ms
   -- GC #2 - Young generation only
   -- Old generation usage went to 36680K out of 61440K, about 60%

GC(3) Pause Initial Mark 37M->37M(79M) 0.414ms
   -- Old generation usage of 60% triggered GC #3, which is a CMS GC

...
GC(3) Concurrent Mark
65   82837504   29176944   53660560
GC(3) Concurrent Mark 2.348ms
   -- GC #3 step ran concurrently

GC(3) Concurrent Preclean
66   82837504   28128352   54709152
67   82837504   27079760   55757744
GC(3) Concurrent Preclean 0.139ms
   -- GC #3 step ran concurrently

GC(3) Concurrent Abortable Preclean

GC(4) ParNew: 17954K->0K(19456K)
GC(4) CMS: 36680K->47944K(61440K)
GC(4) Pause Young (Allocation Failure) 53M->46M(79M) 5.878ms
68   82837504   32693472   50144032
   -- GC #4 (Young GC) performed while GC #3 (Old GC) was still running

...
84   82837504   15546832   67290672
GC(3) Concurrent Abortable Preclean 14.492ms
GC(6) Pause Full (Allocation Failure) 76M->27M(79M) 7.637ms
GC(5) ParNew: 17768K->0K(19456K)
GC(5) CMS: 47944K->28389K(61440K)
GC(5) Pause Young (Allocation Failure) 64M->27M(79M) 14.113ms
   -- GC #3 (Old GC) did not reach the end - cancelled by Full GC #11
   -- GC #6 (Full GC) performed, because heap full
   -- But no details printed for GC #6
   -- GC #5 (Young GC) performed - may be part of GC #6
   -- Not sure why GC #6 printed before GC #5.

...
101   82837504   35448056   47389448
GC(7) ParNew: 17888K->0K(19456K)
GC(7) CMS: 28389K->38632K(61440K)
GC(7) Pause Young (Allocation Failure) 45M->37M(79M) 2.114ms
GC(8) Pause Initial Mark 39M->39M(79M) 0.115ms
GC(8) Concurrent Mark
   -- GC #7 (Young GC) moved more objects to Old generation
   -- That triggered GC #8 (Old GC)

112   82837504   31374912   51462592
GC(8) Concurrent Mark 1.271ms
GC(8) Concurrent Preclean
113   82837504   30326320   52511184
GC(8) Concurrent Preclean 0.115ms
GC(8) Concurrent Abortable Preclean

...
118   82837504   25083360   57754144
GC(9) ParNew: 17768K->0K(19456K)
GC(9) CMS: 38632K->52968K(61440K)
GC(9) Pause Young (Allocation Failure) 55M->51M(79M) 3.210ms
   -- GC #9 (Young GC) performed while GC #8 (Old GC) was still running

GC(8) Concurrent Abortable Preclean 6.264ms
GC(11) Pause Full (Allocation Failure) 69M->29M(79M) 7.138ms
GC(10) ParNew: 17768K->0K(19456K)
GC(10) CMS: 52968K->30437K(61440K)
GC(10) Pause Young (Allocation Failure) 69M->29M(79M) 7.227ms
   -- GC #9 (Old GC) did not reach the end - cancelled by Full GC #11
   -- GC #11 (Full GC) performed, because heap full
   -- But no details printed for GC #11
   -- GC #10 (Young GC) performed - may be part of GC #11

...
   -- Many Old GC started and cancelled by Full GC

...
254   82837504   31376064   51461440
GC(22) ParNew: 17769K->0K(19456K)
GC(22) CMS: 32486K->46824K(61440K)
GC(22) Pause Young (Allocation Failure) 49M->45M(79M) 2.816ms
GC(23) Pause Initial Mark 46M->46M(79M) 0.106ms
GC(23) Concurrent Mark
   -- GC #22 (Young GC) moved more objects to Old generation
   -- That triggered GC #23 (Old GC)

...
266   82837504   21884688   60952816
GC(23) Concurrent Mark 1.429ms
GC(23) Concurrent Preclean
267   82837504   20836096   62001408
GC(23) Concurrent Preclean 0.128ms
GC(23) Concurrent Abortable Preclean
GC(23) Concurrent Abortable Preclean 0.076ms
268   82837504   19787504   63050000
GC(23) Pause Remark 60M->60M(79M) 0.578ms
GC(23) Concurrent Sweep
269   82837504   18738912   64098592
270   82837504   17690320   65147184
GC(23) Concurrent Sweep 0.317ms
GC(23) Concurrent Reset

GC(23) Concurrent Reset 0.097ms
GC(23) Old: 46824K->32488K(61440K)
   -- GC #23 (Old GC) - the first Old GC reached the end!

...

Conclusion:

The Concurrent collector does respect the "-XX:MaxTenuringThreshold=0" setting. Survivor spaces, "from" and "to", will be not used, because all live objects will be promoted directly from "eden" space to "tenured" generation.
The CMS collector starts Old generation GC concurrently and independently from the ParNew collector working on Young generation GC.
When heap is full while an Old GC is running, the Old GC will be cancelled. A Full GC will be performed immediately with the application paused.
The heap parameters were pretty bad. The CMS collector had cancelled many Old GCs and performed Full GCs instead.