Gc algorithms
Download as ODP, PDF5 likes2,220 views
This document provides an overview of garbage collection (GC) algorithms. It begins with an introduction to the speaker and outline. It then discusses why GC is needed, how it detects dead objects using reference tracing and reference counting, and describes basic tracing algorithms like mark-sweep, mark-compact, and copying collection. The document also summarizes generational collection, dynamic GC techniques, multi-threaded GC strategies, and considerations for real-time GC.
![Reference counting (naive)
New():
ref ← allocate()
if ref = null
error “Out of memory”
rc(ref) ← 0
return ref
atomic Write(src, i, ref):
addReference(ref)
deleteReference(src[i])
src[i] ← ref
addReference(ref):
if ref != null
rc(ref) ← rc(ref) + 1
deleteReference(ref):
if ref != null
rc(ref) ← rc(ref) – 1
if rc(ref) = 0
for each fld in Pointers(ref)
deleteReference(*fld)
free(ref)
The Garbage Collection Handbook – Jones, Hosking, Moss](https://image.slidesharecdn.com/gcalgorithms-130706151142-phpapp02/85/Gc-algorithms-30-320.jpg)
Gc algorithms
- 2. Who am I? ● Programming Geek interested in: – GC – JiT Compilers – Concurrency – Non-blocking algorithms – Programming languages runtime
- 3. Outline ● Introduction ● Detecting dead objects ● Basic algorithms ● Generational GC ● Multi-threaded GC ● Real-time GC
- 4. What I'm not covering ● GC tuning ● JVM GC Options ● JVM Specific GC Implementation -Xms8g -Xmx8g -XX:MaxPermSize=256m -XX:NewSize=3G -XX:MaxNewSize=3g -XX:NewRatio=4 -XX:MaxTenuringThreshold=5 -XX:+UseConcMarkSweepGC -XX:+CMSScavengeBeforeRemark -verbose:gc -XX:+PrintGCDetails -XX:+PrintGCTimeStamps -Xloggc:gclogs.txt -XX:+PrintGCApplicationStoppedTime -XX:+PrintGCApplicationConcurrentTime -XX:ParallelGCThreads=7 -XX:+UseGCTaskAffinity -XX:+BindGCTaskThreadsToCPUs -XX:+UnlockDiagnosticVMOptions -XX:ParGCCardsPerStrideChunk=32768
- 5. Why do I need to know about GC? Because GC stops your application!
- 7. Why to collect garbage? ● Limited storage ● Programmers do not like to get dirty ● Programmers make mistakes – Too little collected – memory leaks – error – Too much collected – broken programs – error ● Programmers like good software design – Explicit memory management conflicts with the software engineering principles of abstraction and modularity
- 8. What is garbage? Garbage is an object which does not carry any reference from other objects.
- 9. Detecting dead objects Reference tracing vs Reference counting
- 10. Reference tracing Root references HEAP
- 11. Reference tracing Root references HEAP
- 12. Reference tracing Root references HEAP
- 13. Reference tracing Root references HEAP
- 14. Reference tracing Root references HEAP
- 15. Reference tracing Root references HEAP
- 16. Reference tracing Root references HEAP
- 17. Reference tracing Root references HEAP
- 18. Reference tracing Root references HEAP
- 19. Reference tracing Root references HEAP
- 20. Reference tracing Root references HEAP
- 21. Reference tracing Root references HEAP
- 22. Reference tracing Root references HEAP
- 23. Reference tracing Root references HEAP
- 24. Reference tracing Root references HEAP
- 25. Reference tracing Root references HEAP
- 26. Reference tracing Root references HEAP
- 27. Reference tracing Root references HEAP
- 28. Reference tracing Root references HEAP
- 29. Reference tracing Root references HEAP
- 30. Reference counting (naive) New(): ref ← allocate() if ref = null error “Out of memory” rc(ref) ← 0 return ref atomic Write(src, i, ref): addReference(ref) deleteReference(src[i]) src[i] ← ref addReference(ref): if ref != null rc(ref) ← rc(ref) + 1 deleteReference(ref): if ref != null rc(ref) ← rc(ref) – 1 if rc(ref) = 0 for each fld in Pointers(ref) deleteReference(*fld) free(ref) The Garbage Collection Handbook – Jones, Hosking, Moss
- 31. Reference counting (naive) Root references HEAP
- 32. Reference counting (naive) Root references 1 1 1 2 3 0 1 1 1 1 HEAP
- 33. Reference counting (naive) Root references 1 1 1 1 1 0 1 1 1 1 HEAP
- 34. Reference counting (naive) Root references 1 1 1 1 1 0 1 1 1 1 HEAP Memory leak!
- 35. Detecting dead objects ● Reference tracing ✔ No mutator overhead ✔ Collect cycles ✔ High throughput ✗ Batch style ✗ Not real time ● Reference counting ✔ Incremental ✔ Short pause ✔ Real time ✗ Reference cycles ✗ High mutator overhead ✗ Low throughput
- 36. Basic tracing algorithms Not Moving Moving Mark/Sweep Mark/Compact Copying
- 37. Mark/Sweep vs Mark/Compact Before collection After Mark/Sweep After Mark/Compact Live object Free space Dead object
- 38. Mark/Sweep vs Mark/Compact After Mark/Sweep After Mark/Compact Free list allocation Bump the pointer allocation
- 39. Mark/Sweep vs Mark/Compact ● Mark/Sweep ✔ Fast ✗ Fragmentation ✗ Slower free list allocation ● Mark/Compact ✗ Slow ✔ Compacted heap ✔ Fast bump the pointer allocation
- 40. Copying GC (Ping Pong) ToFrom
- 41. Copying GC (Ping Pong) ToFrom
- 42. Copying GC (Ping Pong) To From
- 43. Copying GC (Survivor spaces) Eden From To Eden From To Eden From To Eden From To
- 44. Copying GC (Survivor spaces) Eden From To Eden To From Eden To From Eden From To
- 45. Copying GC ✔ Compacted heap ✔ The speed depends on the number of live objects ✔ Possible improvement of locality during evacuation ✗ Space overhead
- 46. Generational hypothesis Weak generational hypothesis is the observation that, in most cases, young objects are much more likely to die than old objects. Conversely any object that has survived several GC cycles will probably survive a lot more.
- 47. Generational GC Young space Old space 4 10 16 16 Objects headers Object age Max tenuring threashold = 15
- 48. Generational GC algorithms Young space Old space Copying GC Mark/Sweep Mark/Compact Reference counting
- 49. Dynamic Generational GC G1 GC Eden Survivor Old Humongous Unused
- 50. Card Table and Remembered Set Young space Old space Card Table Dirty Dirty card – write to memory – possible reference to young generation. Will be added to Remembered Set
- 51. Card Table Ref picture: http://blog.ragozin.info/2011/06/understanding-gc-pauses-in-jvm-hotspots.html
- 52. Multi-threaded GC Mutator GC Serial GC Parallel GC Concurrent GC Incremental GC
- 53. Thread Local Allocation Buffer ● Thread allocates within TLAB using bump the pointer ● Improved objects locality ● No contention single pointer TLAB
- 54. Promotion Local Allocation Buffer Each thread has two PLABs: • One for survivor space, • One for tenured space PLAB1 PLAB2 GC Thread 1 GC Thread 2
- 55. Real-time GC Plane: I'm landing. GC: Pff, please wait 2 minutes, I'm collecting
- 56. Real-time GC Real-time systems impose operational deadlines on particular tasks within an application. These real-time tasks must be able to response to application inputs (events) within a fixed time window. The Garbage Collection Handbook – Jones, Hosking, Moss