uschindler commented on PR #15116: URL: https://github.com/apache/lucene/pull/15116#issuecomment-3227093535
Thanks Adrien. It is strange that the overview page of Mike's bench did not show this. It said 0% everywhere. Anyways, the explanation makes sense and confirms that the impact of the group vint is minimal for most queries: we only use group vint for the tail of postings, so only queries with short postings make it visible, otherwise the impact of change is hidden in the large postings that use For posting encoding for the long postings and only the small tail is visible. So in short: let's keep this and backport it? Interestingly the change did not make a difference for Mike's doubt with introducing group vint in vector indexing. I think the allocation profile he has seen was mainly caused by warmup. Lambdas always see a longer warmup time and the allocations caused by the capturing lambda are always seen during warmup until the escape analysis has removed them. Nevertheless it's better without capturing lambda. And code is simpler anyways. In followup I will try to reimplement ByteBuffersDirectory for main branch with a new MemorySegment variant based on current mmap code. But that's unrelated, it will only improve the NRTDirectory. Robert and I had some ideas. Nowadays we can make it also offheap without the garbage collector issues due to missing unmapping. -- 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
