In core compression-How to shrink your database
summary
Demonstrates a Postgres Pro Enterprise page-level compressor (CFS) that writes compressed variable-size pages on eviction, tracks them in a map file and runs segment-based garbage collection; uses ZSTD by default and supports optional encryption via PG_CIPHER_KEY. Synthetic tests show up to 18× smaller files but ~2× slower loads; real workloads show ~5% TPS improvement. Practical tips include 24 bytes of MVCC metadata per row, reordering columns saved ~20% on one DB, and compression is enabled per tablespace.
AI-written summary · about
chapters (23)
- 0:00Introduction
- 1:13Agenda
- 3:05Topol
- 3:39Data size
- 4:38Page inspecting extension
- 5:39Alignment rule
- 6:47Index alignment rule
- 7:22Adding indexes
- 11:36Next release
- 12:07CFS
- 13:21Additional mapping
- 14:20Garbage collection
- 15:02Encryption
- 15:45Usage of compression
- 16:35Disclaimer
- 18:47Real Data
- 19:55Real Date
- 20:07Double buffering
- 20:59Advantages of CFS
- 21:47Cons of CFS
- 23:03Roadmap
- 23:45Questions
- 29:39Conclusion
related talks
32mOn compression of everything in Postgres
28mHow Can Databases Capitalize on Computational Storage?
PolarStore: High-Performance Data Compression for Large-Scale Cloud-Native Databases
46mColumnar Store for PostgreSQL
54mReducing Costs and Improving Performance With Data Modeling in Postgres
14mRe-think Data Management Software Design Upon the Arrival of Storage Hardware with
38mShaving off bytes at any scale: Space savings in various subsystems of PostgreSQL
37mPostgres at Any Scale