Performance Archaeology - 20 years of improvements
summary
Benchmarks run pgbench (Postgres 18 client) and TPC-H on two machines (44‑core/88‑thread for OLTP, SSD small server for OLAP) comparing releases from 8.0→18.0. Key findings: read-only pgbench small dataset: 1.5 GB fits in 2 GB buffer shows huge scalability fixes around 9.4–9.6; OLTP starjoin prepared mode → ~10× higher throughput versus simple mode exposes planning as the limiter; data load for 10 GB falls from ~1 hour on 8.0 to ~15 minutes now (parallel index builds in 11); most OLAP gains cluster around 9.6–11, further wins need execution/storage redesign.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:39Why do this at all?
- 5:38Hardware overview
- 7:38OLTP benchmarks
- 14:03OLTP starjoin benchmarks
- 19:19OLTP summary and future
- 20:30OLAP benchmarks
- 25:55OLAP summary & future
- 26:42The future: what might happen?
- 28:32Resources and conclusion
related talks
50mPerformance Archaeology
31mDemanding the impossible rigorous database benchm
39mPostgreSQL BI Performance & News from XL project
30mOptimising your Database for Analytics
56mPerformance Improvements in PostgreSQL9 2
36mAutomated Performance Testing of Postgres
43mPostgreSQL in the cloud: Theory and Practice
36mBest Practices in PostgreSQL Tuning: Navigating Key Performance Bottlenecks in the Cloud