Automatic database management system tuning through large-scale machine learning

Achieving good performance in DBMSs is non-trivial as they are complex systems with many tunable options that control nearly all aspects of their runtime operation. OtterTune uses machine learning informed by data gathered from previous tuning sessions to tune new DBMS deployments. In experiments with OLTP workloads (on MySQL and Postgres) and OLAP workloads (on Vector), OtterTune produces a DBMS configuration that achieves 58-94% lower latency compared to the default settings or configurations generated by other tuning advisors. We also show that OtterTune generates configurations in under 60 minutes that are within 94% of ones created by expert DBAs. The optimal configuration is different for every application / workload. To demonstrate this the authors take three different workloads and find an optimal configuration for each of them.