Power your JVM with Effective Garbage Collection Tuning

1
Power your JVM with Effective Garbage Collection
Tuning

• SiliconIndia Java Conference
• 29th October, 2010
• Vivekanand Jha

2
Its Science

• MeasuregtUnderstandgtgtTunegtgtMeasuregtgtUnderstandgtgtRet
  une .

3
Agenda

• Some background on JVM GC Architecture
• Demystifying JVM tuning and GC tuning
• Reading the logs

4
(No Transcript)
5
(No Transcript)
6
(No Transcript)
7
Different GC schemes
8
(No Transcript)
9
GC in JDK 7 other exotic GC algorithms

• G1 or Garbage first algorithm, the holy grail of
  Garbage collection?
• Pauseless garbage collection from Azul.
• Garbage collection with hard time guarantee in
  Suns premium product JAVA RTS

10
GC Tuning

• Tuning the Young Generation
• Tuning Parallel GC
• Tuning CMS
• Monitoring GC

11
Dream GC

• Low GC Overhead. ie. High Throughput
• Low GC pause times
• Space efficiency (Less fragmentation)

12
Wisdom droplets

• Supersize it
• Minimize live object for minor collections
• Size the Tenured generation to steady state space
  requirement
• WHEN IN DOUBT, SUPERSIZE IT!!

13
Unnecessary Information

• -Xmxltsizegt max heap size
• ? young generation old generation
• gt -Xmsltsizegt initial heap size
• ? young generation old generation
• gt -Xmnltsizegt young generation size
• gt Applications with emphasis on performance tend
• to set -Xms and -Xmx to the same value
• gt When -Xms ! -Xmx, heap growth or shrinking
• requires a Full GC

14
How to get your object profile in YG

• Monitor tenuring distribution with
• -XXPrintTenuringDistribution
• Desired survivor size 6684672 bytes, new
  threshold 8 (max 8)
• - age 1 2315488 bytes, 2315488 total
• - age 2 19528 bytes, 2335016 total
• - age 3 96 bytes, 2335112 total
• - age 4 32 bytes, 2335144 total

15
Importance of GC Threads
16
Challenges with the CMS
17
CMS-Too early

• ParNew 390868K-gt296358K(773376K), 0.1882258
  secs
• CMS-initial-mark 298458K(773376K), 0.0847541
  secs
• ParNew 401318K-gt306863K(773376K), 0.1933159
  secs
• CMS-concurrent-mark 0.787/0.981 secs
• CMS-concurrent-preclean 0.149/0.152 secs
• CMS-concurrent-abortable-preclean 0.105/0.183
  secs
• CMS-remark 374049K(773376K), 0.0353394 secs
• ParNew 407285K-gt312829K(773376K), 0.1969370
  secs
• ParNew 405554K-gt311100K(773376K), 0.1922082
  secs
• ParNew 404913K-gt310361K(773376K), 0.1909849
  secs
• ParNew 406005K-gt311878K(773376K), 0.2012884
  secs
• CMS-concurrent-sweep 2.179/2.963 secs
• CMS-concurrent-reset 0.010/0.010 secs
• ParNew 387767K-gt292925K(773376K), 0.1843175
  secs
• CMS-initial-mark 295026K(773376K), 0.0865858
  secs
• ParNew 397885K-gt303822K(773376K), 0.1995878
  secs

18
CMS-Too Late

• ParNew 742993K-gt648506K(773376K), 0.1688876
  secs
• ParNew 753466K-gt659042K(773376K), 0.1695921
  secs
• CMS-initial-mark 661142K(773376K), 0.0861029
  secs
• Full GC 645986K-gt234335K(655360K), 8.9112629
  secs
• ParNew 339295K-gt247490K(773376K), 0.0230993
  secs
• ParNew 352450K-gt259959K(773376K), 0.1933945
  secs

19

• -XXExplicitGCInvokesConcurrent
• Requires a post 6 JVM
• -XXExplicitGCInvokesConcurrentAndUnloadClasses

20
Monitoring GC

• You need at least
• ? -XXPrintGCTimeStamps
• (Add -XXPrintGCDateStamps if you must)
• ? -XXPrintGCDetails
• Preferred over -verbosegc as it's more detailed
• gt Also useful
• ? -Xloggcltfilegt
• Separates GC logging output from application
• output

21
Its Art!