Performance-Tuning Twitter Services with Graal and Machine Learning
summary
Twitter runs services on Graal and adds an internal service called Autotune that uses Bayesian optimization to tune three Graal inlining flags (TrivialInliningSize, MaxInliningSize, SmallCompiledLowLevelGraphSize). On the tweet service a 24‑hour Autotune run (44 evaluations) yields ~7% fewer GC cycles and ~6–7% lower CPU vs Graal alone (Graal alone cut CPU ~12% vs hotspot), social‑graph sees ~5% total gain; C2 tuning gives ~5% max, so Graal+Autotune wins.
AI-written summary · about
chapters (13)
- 0:00Start
- 2:35ABOUT ME...
- 6:33PERFORMANCE OPTIMIZATION
- 8:03PERFORMANCE TUNING
- 9:59OPTIMIZATION OF A BLACK BOX FUNCTION
- 14:09WE LIKE BAYESIAN OPTIMIZATION...
- 14:57WHAT IS AUTOTUNE?
- 15:45WHAT IS GRAAL?
- 18:30WHICH PARAMETERS?
- 22:26TEST SETUP
- 23:58TWEET SERVICE: OBJECTIVE
- 24:31TWEET SERVICE: CONSTRAINTS
- 33:28SOCIAL GRAPH: CONSTRAINTS
related talks
49mPerformance tuning Twitter services with Graal and Machine Learning
54mPerformance tuning Twitter services with Graal and Machine Learning
34mPerformance tuning Twitter services with Graal and Machine Learning
56mPerformance tuning Twitter services with Graal and Machine Learning
54mTwitter's quest for a wholly Graal runtime
40mScala ♥ Graal
01h 01mContinuous Optimization of MIcroservices using Machine Learning
22mAutomating Performance Tuning with Machine Learning