antonha commented on PR #13149: URL: https://github.com/apache/lucene/pull/13149#issuecomment-2021556140
> Ideally we'd add a query that adds another implementation of an `IntersectVisitor` to the nightly benchmarks before merging that PR so that we can see the performance bump? Yes - this would be ideal imo. Some caveats that I found while testing: - `PointRangeQuery`s are executed with different `IntersectVisitor`s depending on expected number of matches, which means that several query intervals need to be used in each JVM. In luceneutil, I believe that this corresponds to setting `taskCountPerCat` high. - To see significant change, the number of documents in the segments needs to either fit into the 16bit or 32bit implementation of `bkd.DocIdsWriter` - this was not the case with the simple `localrun.py` script out of the box. - It is possible that we see a performance decrease when we add more `IntersectVisitor` variants, since the benchmark becomes more realistically slow with the additional implementations. -- 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
