SelfTune: Tuning Cluster Managers
summary
SelfTune is a lightweight API plus learning engine that plugs into cluster managers via simple calls like predict and set_reward, uses a continuous bandits algorithm, and tunes real and discrete parameters online with low sample complexity. Deployed to tens of thousands of machines for Microsoft’s background scheduler, it continuously adjusts a concurrency knob to raise throughput up to 20%. In benchmarks it tunes ~100 parameters for DeadStar, cuts rounds-to-good-config from ~30–40 down to about 8, and lowers iteration cost.
AI-written summary · about
chapters (9)
- 0:00Intro
- 0:24Cluster management frameworks
- 4:30Configuring cluster managers
- 7:54Key Features
- 9:08Environments, rewards, optimal configs vary with time
- 9:59Evaluation: Three cluster management settings
- 13:13Azure Function-as-a-Service
- 14:25Hybrid Histogram Policy
- 16:14Summary
related talks
29mContinuous tuning of Microservices via Bayesian Optimization
23mTotal Tidy Tuning Techniques
44mAutomatic parameter tuning for Spark applications
22mAutomating Performance Tuning with Machine Learning
30mScore-Driven Multi-Cluster Management: An Evaluation Framework for…
16mmTuner: Accelerating Parameter-Efficient Fine-Tuning on Multi-GPU Servers with...
35mCreating Your Own LLM Tuning Platform with Open Source Technologies
18mLet the machines optimize the machines
