Concurrent GCs - ZGC & Shenandoah

Java 11 & Java 12 offer two new Garbage Collectors: ZGC and Shenandoah.

The goal of these new concurrent collectors is to reduce GC pauses to only few milliseconds and possibly sub-milliseconds, solving once for all the GC problem.

Transcript

1. sbordet@webtide.com Concurrent GCs ZGC & Shenandoah @simonebordet
2. simone.bordet@gmail.com Simone Bordet • @simonebordet • sbordet@webtide.com • Works @ Webtide ◦ The company behind Jetty and CometD • JVM Tuning Expert 2
3. Introduction 3
4. simone.bordet@gmail.com Introduction Copy Young Generation Old Generation Parallel Mark Compact Stop-The-World Concurrent 4
5. simone.bordet@gmail.com Introduction Copy Young Generation Old Generation Parallel Mark Compact Copy CMS Conc Mark Conc Sweep Stop-The-World Concurrent 5
6. simone.bordet@gmail.com Introduction Copy Young Generation Old Generation Parallel Mark Compact Copy CMS Conc Mark Conc Sweep Copy G1 Conc Mark Compact Stop-The-World Concurrent 6
7. simone.bordet@gmail.com Introduction Copy Young Generation Old Generation Parallel Mark Compact Copy CMS Conc Mark Conc Sweep Copy G1 Conc Mark Compact ZGC Conc Mark Conc Compact Shenandoah Conc Mark Conc Compact Stop-The-World Concurrent 7
8. simone.bordet@gmail.com Introduction • Concurrent GCs ◦ GC runs concurrently with the application • GC races with Application ◦ Marking an object “alive” ◦ Compacting / Moving objects 8
9. simone.bordet@gmail.com Introduction • Concurrent copying example 9 Header a = 1 Header f Region Region
10. simone.bordet@gmail.com Introduction • Garbage Collector compacts object 10 Header a = 1 Header f Header a = 1 Region Region Region
11. simone.bordet@gmail.com Introduction 11 Header a = 1 Header f Header a = 1 Region Region Region "From Space" "To Space"
12. simone.bordet@gmail.com Introduction • Application writes concurrently 12 Header a = 3 Header f Header a = 1 Region Region Region
13. simone.bordet@gmail.com Introduction • Garbage Collector updates references 13 Header a = 3 Header f Header a = 1 Region Region Region LOST WRITE!
14. simone.bordet@gmail.com Introduction • Concurrent GCs store object metadata ◦ E.g. whether an object has been marked • Concurrent GCs require JIT support • JIT injects code that helps the GC ◦ GC barriers 14
15. ZGC 15
16. simone.bordet@gmail.com ZGC • Present in OpenJDK 11+ ◦ Available in AdoptOpenJDK builds ◦ Available in Oracle OpenJDK builds ▪ From https://jdk.java.net • Scalable Low Latency GC • Concurrent Compaction, Single Generation 16
17. simone.bordet@gmail.com ZGC • Only for Linux x86 64-bit ◦ No compressed pointers ◦ ARM port underway • Region Based ◦ ZPages - similar to G1 regions ▪ Small (2 MiB - object size up to 256 KiB) ▪ Medium (32 MiB - object size up to 4 MiB) ▪ Large (4+ MiB - object size > 4 MiB) 17
18. simone.bordet@gmail.com ZGC STW Mark Start GC Thread App Thread 18 Scan threads stacks Mark object roots
19. simone.bordet@gmail.com ZGC Concurrent Marking GC Thread App Thread 19 Mark object graph Gather liveness information
20. simone.bordet@gmail.com ZGC GC Thread App Thread 20 Housekeeping STW Mark End
21. simone.bordet@gmail.com ZGC GC Thread App Thread 21 Weak Reference Processing Free Memory Pages Unload Classes Prepare Relocation Set Concurrent Processing
22. simone.bordet@gmail.com ZGC GC Thread App Thread 22 Relocate Roots STW Relocate Start
23. simone.bordet@gmail.com ZGC GC Thread App Thread 23 Relocate Objects (Compact Heap) Concurrent Relocation
24. simone.bordet@gmail.com ZGC GC Thread App Thread 24 Remapping pointers (Happens During Next Mark) Concurrent Remapping
25. simone.bordet@gmail.com ZGC • ZGC solution ◦ To avoid races with the application • Colored Pointers • Load GC barrier 25
26. simone.bordet@gmail.com ZGC • ZGC colored pointers ◦ Color indicates GC metadata • Check pointer color against GC phase ◦ Wrong color => take action Object address - 44 bits (16 TiB) Unused - 16 bits 64 bits pointer address 4 color bits 26
27. simone.bordet@gmail.com ZGC • ZGC load GC barrier Object f = obj.field; <load_barrier> 27
28. simone.bordet@gmail.com ZGC • ZGC load GC barrier Object f = obj.field; if (addr_of(f) & wrong_gc_color) { slow_path() } 28
29. simone.bordet@gmail.com ZGC • ZGC load GC barrier Object f = obj.field; <load_barrier> mov 0x10(%rdi), %rsi test %rsi, 0x20(%r15) jne slow_path 29
30. simone.bordet@gmail.com ZGC • ZGC load GC barrier tests for the right color • Wrong color -> take slow_path ◦ Fix color & Run some action - atomically • The action depends on the GC phase 30
31. simone.bordet@gmail.com ZGC • ZGC Multi-Mapping Virtual Memory 001<addr> 010<addr> 100<addr> Physical Memory 31
32. simone.bordet@gmail.com ZGC • ZGC Multi-Mapping • RSS shows 3x size ◦ E.g. -Xmx=16G -> RSS~=50G ◦ On Linux, use smem and track PSS • Watch out for OOM Killers in clouds ◦ E.g. AWS 32
33. simone.bordet@gmail.com • Relocation ZGC A E B D C 33 F
34. simone.bordet@gmail.com • Relocation ZGC A E B D F C C’ D’ MetaData B -> B’ C -> C’ D -> D’ 34 B’
35. simone.bordet@gmail.com • Relocation ZGC A E B D F C B’ C’ D’ E’ Load Barrier MetaData B -> B’ C -> C’ D -> D’ E -> E’ 35
36. simone.bordet@gmail.com • Relocation ZGC A E B D F C B’ C’ D’ E’ E” MetaData B -> B’ C -> C’ D -> D’ E -> E’ 36
37. simone.bordet@gmail.com • Relocation ZGC A E B D F C B’ C’ D’ E’ E” MetaData B -> B’ C -> C’ D -> D’ E -> E’ 37 E -> E”
38. simone.bordet@gmail.com • Relocation ZGC A E B D F C B’ C’ D’ E’ F’ MetaData B -> B’ C -> C’ D -> D’ E -> E’ F -> F’ 38
39. simone.bordet@gmail.com • Remapping ZGC A B’ C’ D’ E’ F’ 39
40. simone.bordet@gmail.com ZGC • References ◦ Per Lidén, Erik Österlund et al. ▪ https://www.youtube.com/watch?v=7cWiwu7kYkE ▪ https://www.youtube.com/watch?v=kF_r3GE3zOo ◦ Project Wiki ▪ https://wiki.openjdk.java.net/display/zgc/Main ◦ Mailing List ▪ zgc-dev@openjdk.java.net 40
41. ShenandoahGC 41
42. simone.bordet@gmail.com ShenandoahGC • Present in OpenJDK 12+ repository ◦ Available in AdoptOpenJDK builds ◦ NOT available in Oracle’s OpenJDK builds ◦ Present in RedHat OpenJDK 8, 11+ • Scalable low latency GC • Concurrent Compaction, Single Generation 42
43. simone.bordet@gmail.com ShenandoahGC • Available for x86 32-bit and 64-bit ◦ ARM ports available • Linux, MacOS, Windows ◦ JDK 8, 11 & Latest • Region Based ◦ Derived from G1 43
44. simone.bordet@gmail.com ShenandoahGC STW Mark Start GC Thread App Thread 44 Scan threads stacks Mark object roots
45. simone.bordet@gmail.com ShenandoahGC Concurrent Marking GC Thread App Thread 45 Mark object graph Gather liveness information
46. simone.bordet@gmail.com ShenandoahGC GC Thread App Thread 46 Weak Reference Processing Unload Classes Evacuate Roots Housekeeping STW Mark End
47. simone.bordet@gmail.com ShenandoahGC GC Thread App Thread 47 Evacuate Objects (Compact Heap) Concurrent Evacuation
48. simone.bordet@gmail.com ShenandoahGC GC Thread App Thread 48 Remapping pointers (Happens During Next Mark) Concurrent Update References
49. simone.bordet@gmail.com ShenandoahGC • ShenandoahGC solution ◦ To avoid races with the application • Brooks (Forward) Pointers • Load & Store GC Barriers 49
50. simone.bordet@gmail.com ShenandoahGC • HotSpot Object Header ◦ "Word" == 32 bits on i386; 64 bits on x86-64 Mark Word Class Word Field 1 Field 2 Mark Word Class Word Array Length Slot 1 Slot 2 50
51. simone.bordet@gmail.com ShenandoahGC 1.0 • ShenandoahGC 1.0 Class Field1 Field2 Class Field1 Field2 Mark Mark 51
52. simone.bordet@gmail.com ShenandoahGC 1.0 • ShenandoahGC 1.0 Class Field1 Field2 Class Field1 Field2 Mark Field1 Field2 Mark Field1 Field2 Fwd Ptr Fwd Ptr Mark Mark Class Class 52
53. simone.bordet@gmail.com ShenandoahGC 1.0 • ShenandoahGC 1.0 • Maintains weak to-space invariant ◦ Reads from both from-space to-space copies ◦ Writes only to to-space copy 53
54. simone.bordet@gmail.com ShenandoahGC 1.0 • Load GC barrier: dereference forward pointer Object f = obj.field; <load_barrier> 54
55. simone.bordet@gmail.com ShenandoahGC • Load GC barrier: dereference forward pointer Object f = obj.field; f = deref(addr_of(f) - 8); 55
56. simone.bordet@gmail.com ShenandoahGC 1.0 • Load GC barrier: dereference forward pointer Object f = obj.field; <load_barrier> mov 0x10(%rdi), %rsi mov -0x08(%rsi), %rsi 56
57. simone.bordet@gmail.com ShenandoahGC • Store GC barrier ◦ Runs some action depending on GC phase obj.f = o; <store_barrier> 57
58. simone.bordet@gmail.com ShenandoahGC • Store GC barrier ◦ Runs some action depending on GC phase obj.f = o; if (in_evac_phase && in_collection_set(obj) && !is_forwarded(obj)) slow_path(); 58
59. simone.bordet@gmail.com ShenandoahGC • Store GC barrier ◦ Runs some action depending on GC phase obj.f = o; mov 0x3d8(%r15), %r11 test %r11, %r11 ... jne <slow_path> 59
60. simone.bordet@gmail.com ShenandoahGC • ShenandoahGC 1.0 Problems • More memory needed (due to the forward pointer) ◦ Worst case 50%, common case 5-10% • Complicated (exotic) GC barriers ◦ Load & Store GC barriers to maintain invariants 60
61. simone.bordet@gmail.com ShenandoahGC • ShenandoahGC 2.0 Class Field1 Field2 Class Field1 Field2 Mark Mark 61
62. simone.bordet@gmail.com ShenandoahGC • ShenandoahGC 2.0 Class Field1 Field2 Class Field1 Field2 Mark Field1 Field2 Field1 Field2 Fwd Ptr Mark Mark Class Class Mark Field1 Field2 Class From Space To Space 62
63. simone.bordet@gmail.com ShenandoahGC • ShenandoahGC 2.0 • Maintains a strong to-space invariant ◦ Reads only possible from to-space copy ◦ Writes only possible from to-space copy 63
64. simone.bordet@gmail.com ShenandoahGC 2.0 • Load Barrier (LRB - Load Reference Barrier) ◦ Checks whether there may be forwarded objects ▪ And whether in c-set and whether is forwarded Object f = obj.f; <load_barrier> 64
65. simone.bordet@gmail.com ShenandoahGC 2.0 • Load Barrier (LRB - Load Reference Barrier) Object f = obj.f; if (in_evac_phase && in_collection_set(obj) && !is_forwarded(obj)) slow_path(); 65
66. simone.bordet@gmail.com ShenandoahGC 2.0 • Load Barrier (LRB - Load Reference Barrier) Object f = obj.f; test $0x1,0x20(%r15) ... jne <slow_path> 66
67. simone.bordet@gmail.com ShenandoahGC 2.0 • Shenandoah 2.0 • Load and Store GC barrier are the same! ◦ Big codebase simplification ◦ Improved performance 67
68. simone.bordet@gmail.com • Evacuation ShenandoahGC Class Field1 Field2 Mark 68
69. simone.bordet@gmail.com • Evacuation (GC Copy) ShenandoahGC Class Field1 Field2 Mark Class Field1 Field2 Mark To Space From Space 69
70. simone.bordet@gmail.com • Evacuation (Application triggers LRB) ShenandoahGC Class Field1 Field2 Mark Class Field1 Field2 Mark Class Field1 Field2 Mark To Space From Space 70
71. simone.bordet@gmail.com • Evacuation (Atomic Fwd Ptr Update) ShenandoahGC Class Field1 Field2 Mark Class Field1 Field2 Mark Class Field1 Field2 Mark To Space From Space 71
72. simone.bordet@gmail.com • Evacuation (Application Wins) ShenandoahGC Class Field1 Field2 Fwd Ptr Class Field1 Field2 Mark Class Field1 Field2 Mark To Space From Space 72
73. simone.bordet@gmail.com • Evacuation (Application Read/Write) ShenandoahGC Class Field1 Field2 Fwd Ptr Class Field1 Field2 Mark To Space From Space 73
74. simone.bordet@gmail.com • Update References ShenandoahGC Class Field1 Field2 Fwd Ptr Class Field1 Field2 Mark To Space From Space 74
75. simone.bordet@gmail.com ShenandoahGC • Shenandoah Heuristics ◦ Adaptive (default) ▪ Maintains free heap amount ◦ Static ▪ Starts GC when tripping thresholds ◦ Compact ▪ Small footprint ◦ Aggressive ▪ Back to back GCs ◦ Passive ▪ Non concurrent, only Full GCs ◦ Traversal 75
76. simone.bordet@gmail.com ShenandoahGC • References ◦ Project Wiki ▪ https://wiki.openjdk.java.net/display/shenandoah/Main ◦ Aleksey Shipilëv, Roman Kennke et al. ▪ https://www.youtube.com/watch?v=VCeHkcwfF9Q ▪ https://www.youtube.com/watch?v=E1M3hNlhQCg ◦ Mailing List ▪ shenandoah-dev@openjdk.java.net 76
77. Shenandoah Questions? 77
78. Conclusions 78
79. simone.bordet@gmail.com Conclusions • Give Concurrent GCs a go • Stuck with JDK 8? ◦ Use Shenandoah Already using JDK 11+? ◦ ZGC or Shenandoah • Report your feedback! 79