Performance tuning Twitter services with Graal and Machine Learning
summary
Twitter runs a service that tunes Graal inlining parameters using Autotune (Bayesian optimization as a service) and reports concrete savings. Experiments use dedicated machines, 30-minute evaluations inside 24‑hour runs, tuning three inlining knobs and a single constraint to avoid throttling. Results: switching to Graal cuts user CPU ~13% and GC cycles ~4%, Autotune adds another ~6% for the tweet service (about 18% CPU reduction total) and yields ~5–7% improvements on a social‑graph service; the system is built and produces repeatable gains.
AI-written summary · about
chapters (12)
- 0:00Start
- 1:48ABOUT ME...
- 6:13PERFORMANCE OPTIMIZATION
- 8:46OPTIMIZATION OF A BLACK BOX FUNCTION
- 12:25WHAT IS AUTOTUNE?
- 13:31WHAT IS GRAAL?
- 18:31TEST SETUP
- 20:01TWEET SERVICE: OBJECTIVE
- 20:19TWEET SERVICE: CONSTRAINTS
- 27:31SOCIAL GRAPH: OBJECTIVE
- 30:53WHICH PARAMETERS?
- 35:46SUMMARY
related talks
54mPerformance tuning Twitter services with Graal and Machine Learning
34mPerformance tuning Twitter services with Graal and Machine Learning
46mPerformance-Tuning Twitter Services with Graal and Machine Learning
56mPerformance tuning Twitter services with Graal and Machine Learning
01h 01mContinuous Optimization of MIcroservices using Machine Learning
22mAutomating Performance Tuning with Machine Learning
18mLet the machines optimize the machines
29mContinuous tuning of Microservices via Bayesian Optimization