goankur commented on PR #16145: URL: https://github.com/apache/lucene/pull/16145#issuecomment-6228292193
@gsmiller I ran your `N - 1` variant on f6b3392 next to start-at-`N`, same setup and same session: | | f6b3392 + `N - 1` | f6b3392 + `N` | |---|---|---| | Hot luceneutil, AndHighLow vs main | −3.7% (p = 0.06) | −3.0% (p = 0.11) | | Hot, query types slower / faster (p < 0.05) | 3 / 2 | 9 / 1 | | Cold start, query types slower (arm takes cold JVM / main takes it) | 0 / 0 | 1 / 1 | | Rerank p99, 10 GB cgroup, 2 reps | 17.2 / 17.1 ms | 17.0 / 16.8 ms | Main vs main in this session: 0 slower, 1 faster. For reference, f6b3392 as is was −18% with 15 query types slower. `N - 1` is as good as starting at `N` on hot data and a bit better on a cold start: it catches cold files on the first call and issued about 3.5× more prefetches in the cold JVM. Under memory pressure it matches f6b3392 as is. +1 to merging it with `N - 1`. Data: [followup-gsmiller-n-1](https://github.com/goankur/lucene/tree/pr16145-eval/results/followup-gsmiller-n-1). -- 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]
