Best I/O is No I/O
summary
Claim: minimize I/O by reorganizing data and shifting work out of the database. Uses a composite clustered key (foreign key + timestamp + extra column) to consolidate pages, cutting 30–50 random seeks per page to a few sequential reads and lowering disk busy percent; avoid REPLACE and heavy updates, prefer INSERT/SELECT patterns. Add persistent memcache and deterministic caches in front of NFS, use covering index queries, lean datatypes, server-level connection pools, and consider alternate engines (HBase/MapReduce variants, Tokyo Cabinet) for lower I/O.
AI-written summary · about
chapters (7)
- 0:00Introduction to IO
- 1:36Scaling and system trade-offs
- 3:26Optimizing IO distribution
- 5:55Database indexing strategies
- 8:26Performance testing results
- 12:07Caching and backend tuning
- 14:41Tools and final recommendations
related talks
01h 06mIO in PostgreSQL: Past, Present, Future
34mWhat We Talk About When We Talk About Disk IO
38mScaling IO-bound application in the cloud
48mThe path to using AIO in postgres
41mLinux IO internals for PostgreSQL administrators in 2023
22mI/O in a Flash: Evolution of ONTAP to Low-Latency SSDs
52mHow shit works: Databases
26mLeveraging Moore's Law to Optimize Database Performance