jimczi commented on issue #854: Shared PQ Based Early Termination for Concurrent Search URL: https://github.com/apache/lucene-solr/pull/854#issuecomment-530342313 > Is this proposal to continue collecting the full N per slice (thread work unit), until some/all of them have a full (top N) PQ and at that point you sync to find the "best bottom" across all of them? And every time one of the threads finds a newly competitive hit, increasing its bottom, we sync again to find the best bottom? Yes that's another option that we should investigate imo. > This would still require sync-per-insert, and would require collecting top N per thread instead of top N overall? It's hard to judge which is the better tradeoff w/o building both and running perf tests :) If we keep separate pq we'd not need sync-per-insert and could use a volatile to record the shared min score within the top docs collectors. It seems appealing to me because the top N should be small so the overhead of collecting a full topN per slice is small compared to the converging of the min score which can be completely different per slice.
---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
