Accelerating PostgreSQL Full Table Scan OLAP queries
summary
Tests use PostgreSQL on AWS against a generated 165 million rows / 22 GB dataset and three OLAP queries; a columnar engine cuts disk I/O by about 30% less disk I/O and helps narrow scans when few columns are needed, but can be slower on small scans. A CUDA GPU path (PG strong) achieves up to GPU 400-way parallelism and 10–15% speedups on simple scans but becomes 10–40x slower when multiple joins force data back to the CPU; best results require NVMe + GPU in a single box.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 2:20Overview
- 11:26Possible solutions
- 11:58Storage engines
- 14:42PGStrong
- 15:27Test cases
- 15:53Data model
- 19:45Query types
- 21:06AWS tests
- 21:59Results
- 26:36Second query
related talks
19mFrom OLTP to OLAP: Is PostgreSQL Eating Analytics Too?
30mOptimising your Database for Analytics
34mAnalytics and AI with PostgreSQL - Mission Possible!
50mPerformance Archaeology
29mPerformance Archaeology - 20 years of improvements
10mPostgres on AWS: Run 5X faster on metal, migrate anywhere
21mAnalytics in Postgres –A decade in the making
27mBuilding a PostgreSQL data warehouse