Tuning G1G
summary
G1 is default but not fully hands-off: mis-tuned heaps trigger 15 full GC events in logs, humongous object allocation behavior causes unexpected full GCs, and poor tuning raises operating cost 10% to 20%. Use unified GC logging, Splunk or visualization tools and the provided GitHub parsers to inspect cycles, note region size, IHOP (default 45%), remembered-set churn and kernel-time spikes (~399 hits). Turning on parallel reference processing and adjusting region size or code to avoid humongous allocations often fixes it.
AI-written summary · about
chapters (15)
- 0:00Start
- 3:11OUR MARKETING SLIDE
- 4:09IMPORTANCE OF MODELS
- 5:59ALLOCATORS VS COLLECTORS
- 8:07THINGS WE NEED
- 10:46GIGC ALLOCATION
- 11:58YOUNG COLLECTION
- 17:35GC ROOTS
- 21:22RSET REFINEMENT QUEUE
- 22:34BUILDING A CSET REVISITED
- 24:45SORT BY LIVELINESS
- 25:53MIXED COLLECTION COUNT
- 26:46COLLECTING TENURED
- 28:25HUMONGOUS ALLOCATIONS
- 34:19SOMETHINGS GONE OFF
related talks
59mGC Tuning Confessions of a Performance Engineer
52m-XX:+UseG1GC
39mG1GC Concepts and Performance Tuning
47mGarbage First Garbage Collector
31mDemystifying G1GC
52mGC Tuning & Troubleshooting Crash Course
01h 14mHave you really taken the time to know me: A G1 GC saga
44mMastering GC: tame the beast and make it your best ally