性能工程师指南：玩转OpenJDK-HotSpot垃圾收集器课件.pptx

1、11Performance Engineering2P e r f o r m a n c eE n g i n e e r i n g 2Perfor m a n ce EngineeringPerformance R equirem ents&Test P l a n&DevelopmentPerfor m a n ce Analysis3P e r f o r m a n c e E n g i n e e r i n g P e r f o rRequirementsDesignPerformance R equirem ents&Test P l a n&DevelopmentImp

2、lementationPerfor m a n ce Analysis4R e q u i r e m e n t s D e s i g n P e r f o r m a n c e Perfor m a n ce R equirem ents5P e r f o r m a n c eR e q u i r e m e n t s 56Performance Requirem entsThro ug hput?R espo n se times?Capacity?Fo otpr int?Availability?6 P e r f o r m a n c e R e q u i r e

3、m e n t s T h r o7Performance Analysis7 P e r f o r m a n c eA n a l y s i s8Perfor m a n ce AnalysisMon itor ing System Under TestUtil iz atio n,L o c k Stats,Appl icatio n,System,Memory,ProfilingAnaly zin g logs8 P e r f o r m a n c e A n a l y s i s Mo n i t o r i9JVM Performance Engineering9 J V

4、 MP e r f o r m a n c eE n g i n e e r i n gJava ApplicationJ ava A P IOS+Hardware10JR EJ I T CompilerJava VMG CClass loaderR u ntim eJ a v aA p p l i c a t i o n J a v a A P I O S+Study,Analyz e,TuneJava Devel opm ent KitJ I TC o m piler+RuntimeGarbage Collector/C o l lectio nJ ava API11S t u d y,A

5、 n a l y z e,T u n e J a v a D eGarbage Collection Perfor m a n ce E ng in eer ing12G a r b a g eC o l l e c t i o n P e r f o r m a n cG CPerfor m a n ce EngineeringJVMMonitoringAppl icatio n ProfilingPro cess&Plot G C logsOnline&Off l in e G C MonitoringHeap Prof il ing13G C J V MA p p l i c a t i

6、 o n P r o f i l i n g P r oG arbag e C o l lectio n-F acts,Trade-Offs A n d Algorithms14G a r b a g eC o l l e c t i o n-F a c t s,TG C ca n NOT el im in ate yo ur m em o r y leaks!G C(a n d hea p d um p)ca n pro vid e a n insight into your application.15GC Fact!G Cc a nN O Te l i m i n a t ey o u

7、rm e m oTotal Occupied HeapOld generationYoung generationLive Data Set16T o t a l O c c u p i e d H e a p Y o u n g g e n e rThro ug hput a n d l aten cy are the two m a in drivers towards refinement o f G C algorithms.17GC Fact!T h r o u g h p u ta n dl a t e n c ya r etThe Throughput MaximizerMaxi

8、mize Thro ug hputG enerationalOldParal lel WorkYo ungC o n c ur rent ThreadsStop-t he-wor l d Threads18T h e T h r o u g h p u t Ma x i m i z e r Ma x i m iMr.Latency SensitiveLatency SensitivePa use o n ly when neededConcurrent Mark i ng/SweepingRe d u ce stop-the-wor l d G Cpause timeNo/Partial C

9、o m pactio n19Mr.L a t e n c y S e n s i t i v e L a t e n c ySL ets l o okat Open J DK HotSpot G C s:)20L e t sl o o ka tO p e n J D KH o t S p o tThro ug hput MaximizerLatency Sen sitiveParal lel G C21CMS G CG 1 G CT h r o u g h p u t Ma x i m i z e r L a t e n c y Al l G C s in Open J DK HotSpot

10、are generational.22GC Fact!A l lG C si nO p e n J D KH o t S p o ta r eYo u n g G e n eratio nOld G e n eratio nEd e nSurvivors23Y o u n g G e n e r a t i o n O l d G e n e r a t iEd e nAllocationsSur vivorsYo u n g Generation24E d e n A l l o c a t i o n s S u r v i v o r s Y o u n g25Fast Path All

11、ocation=L o c k-F r e e Allocation=T hread s Al l o cate I nto T heirLo cal Al l o catio n Buff er(LAB)s2 5 F a s tP a t hA l l o c a t i o n=L o cE d e n26TLABTLABTLABTLABTLABTLAB=Thread Local Allocation BufferE d e n 2 6 T L A B T L A B T L A B T L A B T L A B T L A BThread 1Thread 2Thread 3Thread

12、 4TLABTLABEd e nTLABTLABTLABThread 027T h r e a d1 T L A B T L A B E d e n T L A B T L A B T hEd e nAllocationsSur vivorsYo u n g GenerationS0S128E d e n A l l o c a t i o n s S u r v i v o r s Y o u n gEd e nAllocationsSur vivorsYo u n g GenerationS0S129E d e n A l l o c a t i o n s S u r v i v o r

13、 s Y o u n g30G arbag e C o l lectio n-Reclamation.3 0 G a r b a g eC o l l e c t i o n-R e c l a mY o u n g Generation31Old GenerationEd e nS0S1Y o u n gG e n e r a t i o n 3 1 O l dG e n e r a t i*Similar G C Algorithms for Open J DK HotSpotDiff erent G C Al g or ithm s for OpenJDK HotspotAlwa ys

14、co l lected in its entiretyYo u n g Generation32Old Generation*S i m i l a rG C A l g o r i t h m sf o r OAlwa ys co l lected in its entirety33Yo u n g GenerationOld Generation*Similar G C Algorithms for Open J DK HotSpotDiff erent G C Al g or ithm s for OpenJDK HotspotA l w a y sc o l l e c t e di

15、ni t se n t i r eY o ung G arbag e C o l lectio n=Reclamation Via Scavenging34Y o u n gG a r b a g eC o l l e c t i o n=ROldGenerationR o ot SetYo u n g Generation35O l dG e n e r a t i o n R o o tS e t Y o u n gG eSurvivorEd e nY o u n g Generation36S u r v i v o r E d e n Y o u n gG e n e r a t i

16、o n 3 6SurvivorEd e nY o u n g Generation37S u r v i v o r E d e n Y o u n gG e n e r a t i o n 3 7SurvivorEd e nY o u n g Generation38S u r v i v o r E d e n Y o u n gG e n e r a t i o n 3 8Ol d G arbag e C o l lectio n=Different G CAlgorithms39O l dG a r b a g eC o l l e c t i o n=D i fThe Through

17、put CollectorMulti-ThreadedMulti-Threaded-Entire Heap M a r ke d,Swept a n d Compacted in its entiretyAlwa ys co l lected in its entiretyYo u n g Generation40OldGenerationT h e T h r o u g h p u t C o l l e c t o r Mu l t i-Multi-ThreadedMulti-Threaded-Entire Heap M a r ke d,Swept a n d Compacted in

18、 its entiretyAlwa ysco l lected in itsentirety41Yo u n g GenerationOld GenerationMu l t i-T h r e a d e d Mu l t i-T h r e a d e d -Old GenerationR o otSetYo u n g GenerationGarbage Collection-Reclamation via Parallel Mark-Compact42O l dG e n e r a t i o n R o o tS e t Y o u n gG eGarbage Collection

19、-Reclamation via Parallel Mark-CompactOld Generation43G a r b a g e C o l l e c t i o n -R e c l a m a t i oGarbage Collection-Reclamation via Parallel Mark-CompactOld Generation44G a r b a g e C o l l e c t i o n -R e c l a m a t i oGarbage Collection-Reclamation via Parallel Mark-CompactOld Genera

20、tionso urce reg io n45de stinatio n regionG a r b a g e C o l l e c t i o n -R e c l a m a t i oGarbage Collection-Reclamation via Parallel Mark-CompactOld Generationso urce reg io nde stinatio n region46G a r b a g e C o l l e c t i o n -R e c l a m a t i oGarbage Collection-Reclamation via Paralle

21、l Mark-Compactde stinatio n regionsource regionsource regionOld Generation47G a r b a g e C o l l e c t i o n -R e c l a m a t i oGarbage Collection-Reclamation via Parallel Mark-Compactde stinatio n regionsource regionsource regionOld Generation48G a r b a g e C o l l e c t i o n -R e c l a m a t i

22、 oGarbage Collection-Reclamation via Parallel Mark-CompactOld Generation49G a r b a g e C o l l e c t i o n -R e c l a m a t i oThe CMS CollectorMulti-ThreadedMulti-T hreade d-Mostly C o n c ur rent Mark ed a n d Swept in its entiretyAlwa ys co l lected in its entiretyYo u n g GenerationOld Generati

23、on50T h e C MS C o l l e c t o r Mu l t i-T h r e a d eThe CMS CollectorMulti-ThreadedMulti-T hreade d-Mostly C o n c ur rent Mark ed a n d Swept in its entirety?Always collected in its entiretyYo u n g GenerationOld Generation51T h e C MS C o l l e c t o r Mu l t i-T h r e a d eThe CMS CollectorMul

24、ti-ThreadedMulti-T hreade d-Mostly C o n c ur rent Mark ed a n d Swept in its entiretyNo Compaction!Always collected in its entiretyYo u n g GenerationOld Generation52T h e C MS C o l l e c t o r Mu l t i-T h r e a d eGarbage Collection-Reclamation via Mark-SweepOld GenerationF ree List53G a r b a g

25、 e C o l l e c t i o n -R e c l a m a tGarbage Collection-Reclamation via Mark-SweepOld GenerationF ree List54G a r b a g e C o l l e c t i o n -R e c l a m a tA l l non/partial compacting G C s inOpen J DK HotSpot fal lback*to a fu l ly compacting stop-the-world garbage collection called the“full”G

26、 C.55*Tun ing ca n hel p avo id o r po stpo n e f ul l G C s in m a n y cases.GC Fact!A l ln o n/p a r t i a lc o m p a c t i n gG C sSingle Threaded-Entire Heap M a r ke d,Swept a n d Compacted in its entiretyYo u n g GenerationOld Generation56Garbage Collection-Reclamation via Mark-Sweep-CompactS

27、i n g l e T h r e a d e d -E n t i r e H e a pSingle Threaded-Entire Heap M a r ke d,Swept a n d Compacted in its entiretyYo u n g GenerationOld Generation57Garbage Collection-Reclamation via Mark-Sweep-CompactS i n g l e T h r e a d e d -E n t i r e H e a pWhat Abo ut G arbag e F irst G C?58Wh a tA

28、 b o u tG a r b a g eF i r s tG C?5 8Heap RegionsC o ntig uo us J ava Hea pFree Reg io nNon-F ree Region59H e a p R e g i o n s C o n t i g u o u sJ a v aH eThe Garbage First Collector-Regionalized HeapEd e n60OldOldEd e nOldSur vivorHumongousT h e G a r b a g e F i r s t C o l l e c t o r E d eThe

29、Garbage First CollectorEd e n2016 CodeKaram61OldOldEd e nOldSur v ivo rE.g.:C u r rent hea p co n f ig uratio n-T h e G a r b a g e F i r s t C o l l e c t o r E d eThe Garbage First CollectorEd e n2016CodeKaram62OldOldEd e nOldSur v ivo rE.g.:Dur inga yo ung co l lectio n-T h e G a r b a g e F i r

30、s t C o l l e c t o r E d eThe Garbage First CollectorOld2016 CodeKaram63OldOldE.g.:After a yo ungco l lectio n-Sur viv orOl dT h e G a r b a g e F i r s t C o l l e c t o r O l dThe Garbage First CollectorOld2016 CodeKaram64OldOldE.g.:C u r renthea pco n f ig uratio n-Sur viv orOl dEd e nEd e nOldT

31、 h e G a r b a g e F i r s t C o l l e c t o r O l dOld2016 CodeKaram65OldOldSur viv orE.g.:Rec l am atio n o f a g arbag e-f il led reg io n d u r i n g thecleanup phase-Ol dEd e nEd e nOldThe Garbage First CollectorO l d 2 0 1 6C o d e K a r a m 6 5 O l d O l d S u r Old2016 CodeKaram66OldOldSur v

32、iv orE.g.:Rec l am atio n o fag arbag e-f il led reg io n d u r i n g the cleanup phase-Ol dEd e nEd e nThe Garbage First CollectorO l d 2 0 1 6C o d e K a r a m 6 6 O l d O l d S u r Old2016 CodeKaram67OldOldE.g.:C u r rent hea p co n f ig uratio n-Sur viv orOl dEd e nEd e nThe Garbage First Collec

33、torO l d 2 0 1 6C o d e K a r a m 6 7 O l d O l d E.g.Old2016 CodeKaram68OldOldE.g.:Dur ing a m ix ed co l lectio n-Sur viv orOl dEd e nEd e nThe Garbage First CollectorO l d 2 0 1 6C o d e K a r a m 6 8 O l d O l d E.g.Old2016 CodeKaram69E.g.:After a m ix edco l lectio n-Ol dSur vivo rOldThe Garbag

34、e First CollectorO l d 2 0 1 6C o d e K a r a m 6 9 E.g.:A f t eThe Garbage First Collector-Pause HistogramPause time in milliseconds030609012034153416.33417.23418.4341934223423.43432.23433.23436.83437.63438.93440TimestampsYoung CollectionInitial MarkRemark70CleanupMixed CollectionInitiating Heap Oc

35、cupancy PercentT h e G a r b a g e F i r s t C o l l e c t o r P a uF in al ly,L etsTal k Abo ut Tuning!:)71F i n a l l y,L e t sT a l kA b o u t T u nTun ing G CF or T hro ug hput Or For Latency?72T u n i n gG CF o rT h r o u g h p u tO r FG C E l a psed T im e Or G C Overhead73G CE l a p s e dT i

36、m eO rG C O v e r h e aGC Elapsed Timeindicated the amount of time ittakes to executestop the world GC events2016 CodeKaram74T he hig her the G C el a psed tim e-the l o wer the application responsiveness due to the G C induced latenciesG C E l a p s e d T i m e i n d i c a t e d t h e a mOverhead i

37、s an indication of the frequency ofstop the world GC events.2016 CodeKaram75T he m ore f requent the G C events-T he m ore likely it is to negatively impact applicationthroughputO v e r h e a d i s a n i n d i c a t i o n o f t h e f rWhats T he#1 C o ntr ibutor To A G CElapsed Time?76Wh a t sT h e#

38、1C o n t r i b u t o rT oC o pying C o sts!77C o p y i n gC o s t s!7 7What Are T he C o ntr ibutorS To G COverhead?78Wh a tA r eT h eC o n t r i b u t o r ST o Al l o catio n R ate a n d Promotion Rate79A l l o c a t i o nR a t ea n d P r o m o t i o n80Size generations keeping your applications ob

39、ject longevity and size in mind.The faster the generation gets“filled”;the sooner a GC is triggered.Size your generations and age your objects appropriately.The higher the amount of live data to be copied,the longer the GC pause.Premature promotions are a big problem!Tuning Recommendations 8 0 S i z

40、 e g e n e r a t i o n s k e e p i n g y oPlot Allocation&Promotion Rates81P l o t A l l o c a t i o n&P r o m o t i o n RYoung Occupancy before GCYoung Gen SizeOld Gen Occupancy after GC82Heap Occupancy before GCHeap Occupancy after GCHeap SizeTimestampsY o u n g O c c u p a n c y b e f o r e G C Y o u n gQuestions?83Q u e s t i o n s?8 3