gf2121 commented on pull request #541: URL: https://github.com/apache/lucene/pull/541#issuecomment-1012163438
Thanks @iverase ! > I remember in the approach I tried I was batching the docIds by a number (128 or 256) but in general tI didn't see much better performance comparing to the added complexity. I saw the benchmark result in https://github.com/apache/lucene-solr/pull/1538. It is similar to the benchmark result mentioned above: around 20% QPS increase for 1D high cardinality cases. But it seems like the benchmark did not cover all use cases. Try to think about some worst cases, e.g. low cardinality fields (It would be great if you can take a look at [here](https://github.com/apache/lucene/pull/541#issue-1080062490)). I proposed a BitSet approach several days ago to solve this but that's not enough since the optimization can hardly be triggered when cardinality comes to 32+. > I don't like that the optimisation only works for a specific number of points, it feels trappy. +1, I made a [change](https://github.com/apache/lucene/pull/541/commits/ccdae2f6920abff52abf6aae42d65b33861d4cf6) to make the BKDForutil flexible and benchmark seems the performance doesn't show a significant regression. -- 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: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
