Workload-Based MySQL Tuning
summary
A workload-aware tuning pipeline maps a target workload to a closest past run using DBMS internal metrics, prunes redundant signals with factor analysis + k-means, and prompts an OpenAI LLM to propose 1–3 candidate knob sets that pass version and resource checks. An agent applies configs, gathers runtime and external metrics, and the loop picks the next config by maximizing estimated gain. Experiments on bench workloads reach best results within 20 iterations and report up to 54% lower latency.
AI-written summary · about
chapters (9)
- 0:00Workload-based MySQL tuning intro
- 3:17Manual MySQL config challenges
- 5:51LLM-based tuning data reuse
- 8:47Characterizing workload with DBMS metrics
- 11:20Using large language models for tuning
- 13:52Generating configs with LLM
- 15:36Experiment setup & data
- 16:40Results & performance gains
- 17:25Q&A
related talks
25mDatabase Parameter Auto-Tuning Through CI/CD Pipelines
53mIntroduction to MySQL Performance Tuning - workshop 1hr
53mMySQL DevOps
18mLet the machines optimize the machines
14mStop Hand Tuning MySQL Teach Kubernetes Operator
26mHow MySQL Configuration Tuning Boosts App Performance
57mPerformance Analysis and Optimization of MySQL
43mOptimizing MySQL without SQL or touching my.cnf