Continuous tuning of Microservices via Bayesian Optimization
summary
Autotune uses Bayesian optimization backed by White Lab to tune JVM parameters (typically 20–50 knobs) by running controlled experiments on production instances, comparing a control and an experimental instance, and rolling back on constraint violations via the observability system. It integrates with Aurora to deploy configs, logs experiments to storage, and reports 7–8% improvements on major services. The system is live as Autotune, handles staging/production differences by running in prod, and still needs better noise handling and stopping criteria.
AI-written summary · about
chapters (4)
related talks
41mAutomated Performance Tuning with Bayesian Optimization
01h 01mContinuous Optimization of MIcroservices using Machine Learning
34mPerformance tuning Twitter services with Graal and Machine Learning
54mPerformance tuning Twitter services with Graal and Machine Learning
46mLinux Kernel Autotuning
22mAutomating Performance Tuning with Machine Learning
18mLet the machines optimize the machines
56mPerformance tuning Twitter services with Graal and Machine Learning