Boosting the Priority of Garbage: Scheduling Collection on Heterogeneous Multicore Processors

Transcription

1 Boosting the Priority of Garbage: Scheduling Collection on Heterogeneous Multicore Processors Shoaib Akram, Jennifer B. Sartor, Kenzo Van Craeynest, Wim Heirman, Lieven Eeckhout Ghent University, Belgium

2 Popularity of Managed Languages The 215 Top Ten Programming Languages, spectrum.ieee.org. 2

3 The Garbage Collection Advantage Memory automatically reclaimed for reuse Takes extra CPU cycles to provide the service 3 Concurrent collectors suited to multicores

4 Heterogeneous Multicores Performance LITTLE 6 Series 4x ARM Cortex A72 4x ARM Cortex A53 In-Order big Out-of-Order Exynox 889 4x ARM Cortex A53 4x Exynos M1 Power 4

5 Managed Language Applications on Heterogeneous Multicores Performance Application à big Out-of-Order LITTLE In-Order Power Garbage Collector à big or LITTLE? 5

6 GC on big versus LITTLE Applica;on and collector running concurrently Applica'on Allocates objects on heap big Collector Iden;fies live objects on heap and then reclaims memory taken up by remaining objects big LITTLE Run Collector on big versus LITTLE and measure the difference in execution time 6

7 GC on big versus LITTLE % increase in execution time

8 GC on big versus LITTLE % increase in execution time

9 GC on big versus LITTLE % increase in execution time GC-Uncritical GC-Critical Some applications exhibit GC-Criticality GC on LITTLE detrimental for GC-Critical 9

10 GC on big versus LITTLE What happens if GC runs on LITTLE for GC-Cri;cal apps? Applica'on Allocates objects on heap Paused!!! Collector Iden;fies live objects on heap and then reclaims memory taken up by remaining objects Serial collec;on Application is paused if no free memory on heap because collector still running 1

11 Giving GC Fair Share of Big Core gc-fair Equally share the big core among all threads Based on Van Craeynest et al [PACT 213] Baseline is gc-on-little Pin the GC threads on LITTLE cores Observe the % reduc;on in execu;on ;me 11

12 Giving GC Fair Share of Big Core % execution time reduction LITTLE 3 LITTLE 1 LITTLE GC-Uncritical 12

13 Giving GC Fair Share of Big Core % execution time reduction LITTLE 3 LITTLE 1 LITTLE GC-Uncritical 13

14 Giving GC Fair Share of Big Core % execution time reduction LITTLE 2 LITTLE 1 LITTLE GC-Uncritical gc-on-little for GC-Uncritical 14

15 Giving GC Fair Share of Big Core % execution time reduction LITTLE 2 LITTLE 1 LITTLE GC-Uncritical GC-Critical gc-on-little for GC-Uncritical 15

16 Giving GC Fair Share of Big Core % execution time reduction LITTLE 2 LITTLE 1 LITTLE GC-Uncritical GC-Critical gc-on-little for GC-Uncritical 16

17 Giving GC Fair Share of Big Core % execution time reduction LITTLE 3 LITTLE 1 LITTLE GC-Uncritical GC-Critical gc-on-little for GC-Uncritical gc-fair for GC-Critical 17

18 Giving GC Fair Share of Big Core % execution time reduction LITTLE 2 LITTLE 1 LITTLE GC-Uncritical GC-Critical GC-Criticality depends on architecture, application, and runtime environment 18

19 Our Contribution 25 3 LITTLE LITTLE GC-Criticality-Aware 1 LITTLE 1 % execution time reduction GC-Uncritical Scheduler GC-Critical Dynamically adjusts # big core cycles given to the concurrent collector GC-Criticality depends on architecture, application, and runtime environment 19

20 GC-Criticality-Aware Scheduler Runtime Activity à How Scheduler Reacts? app gc Schd. App alone gc-on-little 'me 2

21 GC-Criticality-Aware Scheduler gc-on-little to gc-fair app gc Schd. App alone gc-on-little 'me 21

22 GC-Criticality-Aware Scheduler gc-on-little to gc-fair JVM signals the scheduler app gc Schd. App alone Stop Concurrent gc-on-little Scan gc-fair 'me Stop pause to do book-keeping ignored Scan stop pause: JVM signals scheduler gc-fair gives equal priority to GC and app 22

23 GC-Criticality-Aware Scheduler Boost States Stop scan pauses observed even with gc-fair Scheduler Scheduler How many quanta scheduled on the BIG core? gc-on-little First GC thread =, Second GC thread = gc-fair First GC thread = 1, Second GC thread = 1 Boost the priority of garbage Give GC more consecu;ve quanta on big State How many quanta scheduled on the BIG core? gc-boost P First GC thread = 1, Second GC thread = 1 gc-boost P1 First GC thread = 1, Second GC thread = 2 Degrade boost state when no longer cri;cal 23

24 GC-Criticality-Aware Scheduler gc-boost:p to gc-on-little JVM signals the scheduler App alone Stop Concurrent App alone 'me app gc Schd. gc-boost:p gc-on-little If no scan pause in state P, go to gc-on-little Can configure # zero stop scan intervals before returning to gc-on-little 24

25 GC-Criticality-Aware Scheduler Summary JVM detects GC-Criticality during runtime Communicates criticality information down to the scheduler Scheduler dynamically adapts big core cycles given to GC 25

26 Experimental Setup Java Virtual Machine Jikes Research Virtual Machine (Version 3.1.2) Full-heap concurrent collector with two threads Tackle non-determinism by warming up the JVM Heap size 2x of minimum Benchmarks Ten benchmarks from DaCapo Vary the # threads 1 to 4 Heterogeneous Multicore Setup Sniper multicore simulator (Version 4.) Different four core heterogeneous architectures Varying # of big and LITTLE cores 26

27 Performance of GC-Criticality-Aware Scheduler % execution time reduction big plus one LITTLE core gc-fair GC-Uncritical GC-Critical 27

28 Performance of GC-Criticality-Aware Scheduler % execution time reduction big plus one LITTLE core gc-fair gc-boost GC-Uncritical GC-Critical gc-boost performance neutral for GC-Uncritical 28

29 Performance of GC-Criticality-Aware Scheduler % execution time reduction big plus one LITTLE core gc-fair gc-boost GC-Uncritical GC-Critical gc-boost performance neutral for GC-Uncritical 29 Improves perf. of GC-Critical by 14% on avg.

30 Understanding the Performance Advantage of Big Core Cycles per instruction Application Collector L3 Miss L2 Miss L1-D Miss L1-I Base 3

31 Understanding the Performance Cycles per instruction Advantage of Big Core LITTLE Application Collector L3 Miss L2 Miss L1-D Miss L1-I Base Collector performs a heap traversal chasing pointers 31

32 Understanding the Performance Cycles per instruction Advantage of Big Core LITTLE big Application Collector L3 Miss L2 Miss L1-D Miss L1-I Base Collector performs a heap traversal chasing pointers Instruction-level parallelism J Memory-level parallelism L 32

33 Performance of GC-Criticality-Aware Scheduler % execution time reduction Lowering frequency of LITTLE core Similar freq. GC-Uncritical GC-Critical 33

34 Performance of GC-Criticality-Aware Scheduler % execution time reduction Lowering frequency of LITTLE core Similar freq. 1 GHz slower GC-Uncritical GC-Critical Lowering frequency increases GC-Criticality 34

35 Performance of GC-Criticality-Aware Scheduler % execution time reduction Lowering frequency of LITTLE core Similar freq. 1 GHz slower GC-Uncritical GC-Critical Lowering frequency increases GC-Criticality Improves perf. of GC-Critical by 2% on avg. 35

36 % execubon Bme reducbon Different # LITTLE cores Performance of GC-Criticality-Aware Scheduler GC-UnCri;cal GC-Cri;cal 1L 2L 3L Allocation rate lowers with more LITTLE cores gc-boost is beneficial for different # LITTLE 36

37 % reduction in energy-delay product Energy Efficiency of GC-Criticality-Aware Scheduler big plus one LITTLE core GC-Uncritical GC-Critical Negligible change in EDP for GC-Uncritical 2% avg. reduction in EDP for GC-Critical 37

38 More in the Paper Sensitivity studies Varying number of total cores Scheduling quantum and # zero scan intervals Heap size GC-Criticality using OpenJDK s collector 38

39 Conclusions Concurrent garbage collection benefits from out-of-order execution Java applications that allocate rapidly exhibit GC-Criticality GC-Criticality-Aware scheduler adjusts big core cycles given to GC on a heterogeneous multicore Uses information provided by the JVM Improves both performance and energy efficiency 39

40 Boosting the Priority of Garbage: Scheduling Collection on Heterogeneous Multicore Processors Thank You!

41 GC Criticality with OpenJDK s CMS 8 % increase in execution time

42 Triggering Concurrent GC Every 32 MB of Allocation % reduction in energy delay product

43 GC-Criticality-Aware Scheduler gc-boost:p to gc-boost:p1 JVM signals the scheduler App alone Stop Concurrent Scan 'me app gc Schd. gc-boost:p gc-boost:p1 gc-boost:p1 gives GC two quanta on big 43

44 GC-Criticality-Aware Scheduler gc-boost:p1 to gc-boost:p JVM signals the scheduler App alone Stop Concurrent App alone 'me app gc Schd. gc-boost:p1 gc-boost:p Degrade boost state if no stop scan pause 44

45 % reduction in energy-delay product Energy Efficiency of GC-Criticality-Aware Scheduler big plus one LITTLE core 45

46 % reduction in energy-delay product Energy Efficiency of GC-Criticality-Aware Scheduler big plus one LITTLE core GC-Uncritical Negligible change in EDP for GC-Uncritical 46