The Alchemy of Shared Buffers: Balancing Concurrency and Performance

*Josef Machytka: Speaker portfolio*

This talk offers source code guided dive into PostgreSQL’s shared buffers architecture: the three‑layer design, the hit/miss/I‑O lifecycle, and how pins, per‑page LWLocks, and atomic BM_* flags coordinate page‑granular concurrency. We will trace clock‑sweep and buffer rings that prevent big scans and VACUUM from polluting the cache.

Key takeaways

• Architecture & synchronization: three-layer design; buffer pins + per-page LWLocks + partitioned hash table mapping enable fine-grained concurrency
• Eviction & cache pollution control: clock-sweep algorithm maintains hot pages in cache; buffer rings fence off bulk operations like sequential scans and VACUUM
• Performance tuning: balance double-buffering effects, size shared_buffers sensibly, optimize checkpoint parameters; validate with EXPLAIN (BUFFERS) and pg_stat_bgwriter metrics

Presented At & Slides

• CERN PGDay 2026 (slides — talk 45 minutes)
• HOW2026 Jinan China (slides — talk/workshop 3 hours, repository)

Recording

[embed]The Alchemy of Shared Buffers: Balancing Concurrency and Performance Talk dissects PostgreSQL's shared buffers: the three‑layer design (buffer pool → BufferDesc headers → buf_table hash)…videos.cern.ch

Photos

© HOW2026 conference orginizers

© Adrien Obernesser — LinkedIn