JingsongLi opened a new pull request, #57: URL: https://github.com/apache/paimon-vector-index/pull/57
## Summary Optimizes IVF-RQ search hot paths without changing public APIs or the on-disk index format. ## Changes - Build the `query_bits=0` byte LUT incrementally. - Replace linear Top-K worst-result scans with an indexed max heap. - Partially select IVF coarse centroids and sort only the `nprobe` prefix. - Batch RQ reader multi-range reads with a 16 MiB payload bound. ## Performance ANN benchmark (`N=10,000`, `NQ=1,000`, `D=128`, `nlist=64`, `nprobe=8`, `k=10`, `query_bits=0`): - Search latency: 130 ms -> 86 ms - Throughput: 7,673 -> 11,627 QPS (~1.5x) ## Testing - `cargo test --workspace` - `cargo clippy --workspace --all-targets -- -D warnings` - `cargo fmt --check` - `cargo bench -p paimon-vindex-core --bench ann_bench -- --nocapture` ## Notes No breaking changes or storage-format migration required. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
