IO in Postgres - Architecture, Tuning, Problems
Presented by:
Andres Freund
Andres is a PostgreSQL committer and developer, focusing on replication, scalability and robustness. Before Citus Data he worked as a PostgreSQL Developer and Consultant at 2ndQuadrant and as a freelancing consultant in the areas of databases and software engineering. He has been developing Postgres and other Open Source projects since 2005. In his free time he enjoys climbing, diving and reading paper books.
No video of the event yet, sorry!
To get the most out of PostgreSQL when used in an I/O intensive environment will often require some tuning. Aside from a few basic guidelines, tuning requires some understanding of how Postgres does I/O, how it caches data and so forth. Among others we'll go through the architecture of the buffer cache ('shared_buffers'), how checkpointing (checkpoint_segments) influences performance, what the wal writer and background writer do; and how to tune each of those. While hopefully most of the time staying above the too nitty gritty details of how exactly things are implemented. In addition there are some recognized problems with postgres in that area - I'll try to describe which those are, and how those might be/are getting fixed.
- Date:
- Duration:
- 30 min
- Room:
- Conference:
- PGConf US 2016 [PgConf.US]
- Language:
- Track:
- Internals
- Difficulty:
- Medium