Make Your Database Dream of Electric Sheep: Designing for Autonomous Operation
summary
Research prototypes show concrete gains: auto-tune using Gaussian processes on GPflow/TensorFlow automatically recommends knob sets for Postgres/MySQL, climbing a demo Postgres run from ~623 tps to ~2,600 tps and matching or beating expert DBAs on several benchmarks. A second system, Peloton with forecasting plus receding-horizon planner, embeds workload forecasting (deep learning) and Monte‑Carlo search to plan index/scale actions. Design rules: mark human-value knobs off-limits, expose hardware and per-component metrics, avoid restarts, and allow controlled divergent replicas for safe exploration.
AI-written summary · about
chapters (30)
- 0:00Introduction
- 1:35Why Autonomous Systems
- 2:50SelfAdaptive Databases
- 8:40Why is this all insufficient
- 10:35Why are things different now
- 17:55Autotune demo
- 22:27Peloton
- 28:45Demo
- 29:42Overview
- 29:55Agenda
- 30:21PG Settings
- 30:56Metadata
- 31:27Deltas
- 32:49Models
- 33:35Metrics
- 34:22tunable components
- 34:39column families
- 34:59tuning column families
- 35:49why this matters
- 36:58notifications
- 37:38Oracle
- 37:56Replicating
- 38:20Replication Example
- 39:43Oracle Rant
- 40:07We had this idea
- 40:17Andy sent Larry an email
- 40:35Key ideas
- 43:02Final thoughts
- 43:53Results
- 45:04Questions
related talks
01h 00mPostgreSQL and Artificial intelligence
50mLeveraging AI as a PostgreSQL DBA
33mThe Building Blocks for Self-Driving PostgreSQL
48mChatGPT Ain’t Got $%@& On Me! Next Generation Automated Database Tuning
45mPostgres and the Artificial Intelligence
37mRound Table Discussion
46mPower the next generation of AI apps with databases at scale anywhere
20mOpen Source Database Operations in the Age of AI