Downloaded 19 times
Uploaded byAtlassian
Trash Talk! How to Reduce Downtime by Tuning Garbage Collection
Denise Unterwurzacher discusses strategies for reducing downtime in JVM through effective garbage collection tuning. The document covers garbage collection basics, troubleshooting methods, and best practices for optimizing performance without causing adverse effects. Key takeaways include the importance of benchmarking and having a systematic approach to tuning JVM parameters for better application performance.
More Related Content
Similar to Trash Talk! How to Reduce Downtime by Tuning Garbage Collection
Trash Talk! How to Reduce Downtime by Tuning Garbage Collection
- 1. DENISE UNTERWURZACHER •CONFLUENCE ENGINEER • ATLASSIAN Trash Talk! How to Reduce Downtime by Tuning Garbage Collection
- 2. GARBAGE COLLECTION BASICS ACAUTIONARY TALE TROUBLESHOOTING Agenda TUNING BEST PRACTICES
- 3. The defaults are sane TheJVM is built to react adaptively The knock-on effects can be hard to predict Don’t mess with it
- 4. EAC - extranet.atlassian.com Node3Node 2Node 1 5.5gb VM 5.5gb VM 5.5gb VM
- 5. EAC - extranet.atlassian.com 1250daily users
- 6.
- 7.
- 8.
- 9.
- 10.
- 11.
- 12.
- 13.
- 14. ‘Stop the world’- stops the application running to collect garbage Collects garbage while the application runs Serial/Parallel Concurrent
- 15. Single or multi- threaded Serial/Parallel Goodfor small heaps ‘Stop the world’ full GCs
- 16. G1GC Needs a largerheap Still some pauses No full GCs
- 17. Young generation collections arenever concurrent.
- 18.
- 19. Pause types To-space allocation FullGC Young generation Mark/remark/cleanup Serial/parallel, concurrent failover All collectors Concurrent Concurrent
- 20.
- 21. Benchmarking Number of pauses Lengthof pauses Time between pauses
- 22.
- 23. Garbage collection is hard Fixescan make things worse Benchmarking is essential Lessons learned
- 24.
- 25.
- 26.
- 27.
- 28.
- 29.
- 30.
- 31.
- 32.
- 33.
- 34.
- 35.
- 36.
- 37.
- 38.
- 39.
- 40.
- 41.
- 42.
- 43.
- 44.
- 45.
- 46.
- 47.
- 48.
- 49.
- 50.
- 51.
- 52.
- 53.
- 54.
- 55.
- 56.
- 57. OutOfMemory Error -Java Heap Space
- 58.
- 59. Bad tuning -new generation too small
- 60. Heap too large GCtimes
- 61. Long pauses -allocation failure To-space allocation failure Full GC
- 62. Long pauses -unable to free enough space
- 63. Solutions Long pauses Heap toolarge OOMEs Increase memory by small increments. Decrease memory. Diagnose, benchmark, test.
- 64.
- 65.
- 66. Tuning Goals Footprint Throughput Latency Maximum GCpause times Time spent serving application requests The size of the heap
- 67. Footprint Throughput Pause times Low pausetimes + high throughput Use G1GC and a large heap
- 68. Footprint Throughput Pause times Low pausetimes + low footprint Use G1GC and a medium sized heap
- 69. Footprint Throughput Pause times Low pausetimes + low footprint Use G1GC and a medium sized heap
- 70. Footprint Throughput Pause times High throughput+ low footprint Use a parallel collector and a small heap
- 71. Footprint Throughput Pause times High throughput+ low footprint Use a parallel collector and a small heap
- 72. JVM auto-tuning 3. Footprint 2. Throughput 1.Latency Maximum GC pause times Time spent serving application requests The size of the heap
- 73. Overtuning Increasing the heap toomuch Parameters that stop the heap auto- tuning
- 74. -XX:MaxNewSize=2g -XX:NewRatio=3 Hard tuningSoft tuning Young Generation
- 75. -XX:GCTimeLimit=10 -XX:GCTimeRatio=9 Hard tuningSoft tuning -XX:MaxGCPauseMillis=500 Throughput
- 76. -XX:SurvivorRatio=4 -XX:G1ReservePercent=20 Hard tuningSoft tuning Survivor
- 77.
- 78. If you haveto tune, tread carefully.
- 79.
- 80.
- 81. What you can dowhen you get home Be minimal! Think about goals Review current state
- 82. Where to get helpwith GC Atlassian Support Oracle’s docs Atlassian’s docs https://confluence.atlassian.com/x/ aACCGw http://tinyurl.com/caq6y7h https://support.atlassian.com
- 83. Thank you! DENISE UNTERWURZACHER• CONFLUENCE ENGINEER • ATLASSIAN