viliam-durina commented on PR #15590: URL: https://github.com/apache/lucene/pull/15590#issuecomment-3809609368
TL;DR: I'm no longer claiming this PR gives 15% improvement to parallel HNSW graph building. It's only a race fix and perhaps a small performance improvement. So it's sufficient to backport to 10.4. I saw 15% slowdown in one particular dataset when upgrading from 10.2.1 to 10.2.2. It was 15% on AWS instance, but only 5% on my laptop (with performance and efficient cores and thermal throttling). However I'm using a Lucene fork and there are other non-trivial changes involved. I was investigating my changes and couldn't find the cause, and I found the issue addressed in this PR, and on my laptop I saw a few % improvement, so I claimed 15% improvement. However, when I was later trying to reproduce the improvement on an AWS instance, I couldn't observe measurable improvement, as each benchmark run is a few % different. -- 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]
