neoremind commented on issue #16574: URL: https://github.com/apache/lucene/issues/16574#issuecomment-5865343968
@rmuir @mikemccand would love to get your thoughts on this when you have time. I've laid out the benchmarks with both micro and end-to-end w/ perf and analysis above, along with my thoughts and concern. In short, it is not convincing in microbenchmarks with a slight regression, but shows a consistent good end-to-end win on real Wikipedia indexing due to inter-doc cache pollution, better frequent-term cache locality, and the sort phase benefits the most regardless of data traits. On the other side, there's a slight regression on low-frequency terms as co-location of term bytes and the first postings slice is good for that case. I tend to keep as-is if we don't want to compromise some workloads a bit while being advantageous (even substantial here) to others. -- 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]
