The goal wasn’t to build a general database but to explore how far a minimal, modern B+Tree design can go when tuned for CPU-cache behavior, small fixed-size pages, and predictable read/write paths.
In my benchmarks (linked in the repo), BTreePlus shows better throughput than SQLite and Postgres for specific key-value style workloads—mainly:
- single-key point lookups
- sequential inserts with small pages
- read-heavy scenarios without complex SQL layers
I’d love feedback from database and systems engineers on:
- the page layout / split-merge logic
- the latching model
- whether the design avoids classic B-Tree pitfalls
- benchmarking methodology (happy to adjust or rerun)