Utkarsh-rwt commented on issue #16780: URL: https://github.com/apache/lucene/issues/16780#issuecomment-6125960562
Hi! I've completed a standalone benchmark comparing Lucene's existing `FuzzySet` Bloom filter with FastFilter Java's Binary Fuse 8-bit and 16-bit filters. I audited the benchmark, corrected the query-hashing comparison, validated the raw and summary CSVs, and verified correctness and serialization checks. No Lucene production code was modified. For 1 million keys, the results showed: - **Bloom (1% target):** 2,097,161 bytes, 132.4 ms construction time, measured FPR of 0.0306%. - **Binary Fuse 8-bit:** 1,130,516 bytes, 76.6 ms construction time, measured FPR of 0.3847%. - **Binary Fuse 16-bit:** 2,261,012 bytes, 92.2 ms construction time, measured FPR of 0.0011%. These are standalone filter measurements, not end-to-end Lucene search results. Binary Fuse 8-bit appears promising in terms of serialized size and construction time, while the 16-bit variant achieves a lower measured FPR at a higher storage cost. The next step I'm considering is a real Lucene index benchmark covering segment writing, merging, memory overhead, and present/absent exact-term lookups. Would these preliminary results and a follow-up integration benchmark be useful for the issue? [environment.txt](https://github.com/user-attachments/files/33341543/environment.txt) [raw-results.csv](https://github.com/user-attachments/files/33341544/raw-results.csv) [summary.csv](https://github.com/user-attachments/files/33341545/summary.csv) -- 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] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
