CAPI/OpenCAPI Accelerated PostgreSQL
summary
Implementation accelerates PostgreSQL string queries by adding FPGA logic connected over a cache‑coherent CAPI/OpenCAPI bridge and using the SNAP framework. Regular‑expression engine is written in RTL with 8 engines × 16 pipelines running ~225 MHz, host software uses multithreaded workers, packet/result buffers and AXI fetches via virtual addresses. Testbed is 2×22‑core CPUs with 512 GB RAM; measured 10x to 2x overall throughput vs. a CPU build. Remaining work: hardware job manager and direct page‑pointer access.
AI-written summary · about
chapters (13)
- 0:00Introduction to CAPI/OpenCAPI
- 1:10Acceleration overview and interfaces
- 2:17System architecture details
- 5:46Hardware engine specifics
- 7:14Software implementation with hooks
- 9:45Multi-threading worker design
- 11:31Live demo walkthrough
- 13:46Performance evaluation setup
- 16:15Speed-up results vs CPU
- 17:40Thread count impact analysis
- 19:29Job count and cleanup optimization
- 21:14Future work and enhancements
- 23:42Conclusion and Q&A
related talks
56mPerformance Improvements in PostgreSQL9 2
35mAccelerating PostgreSQL Full Table Scan OLAP queries
52mWriting fast C code for a modern CPU (and applying it to PostgreSQL)
06mAchieve predictable performance
34mAnalytics and AI with PostgreSQL - Mission Possible!
50mFrom Transactions to Intelligence: Postgres for Open AI & Analyti
34mAnalytics and AI with PostgreSQL - Mission Possible!
17mHardware Accelerated Big Data Processing