Making Postgres 300x faster for analytics: batching, operator fusion, and SIMD
Points and comments are a snapshot, not live.
A Rust-based Postgres extension claims 300x speedups for analytics via batching, operator fusion, and SIMD.
The pgrust project rebuilt Postgres's query engine in Rust, targeting CPU-bound analytical workloads. Benchmarks show a simple sum query dropped from ~20s in Postgres to 135ms after applying batching (1024-row buffers), operator fusion to eliminate data copies, and SIMD accumulation on ARM. The changes yield a 9.6x improvement over the Volcano-model baseline and a 300x gain over stock Postgres on ClickBench. The author notes Postgres's age (1980s design for disk-bound workloads) as the root cause of its inefficiency. The project is AGPL-licensed.
What commenters are saying
License concerns dominate the thread. Many commenters call AGPL a dealbreaker for corporate use, arguing it blocks adoption by the large customers who would benefit most. The author counters that AGPL is standard for databases today and offers commercial dual-licensing. Skeptics question the project's credibility, noting that early commits are AI-generated (Claude) and that GitHub stars are easy to fabricate. A defender points to 6000 commits on the v0.2 tag and the author's formal verification work, which has found 20 bugs in Postgres itself.