[GitHub] [lucene] uschindler commented on issue #12527: Optimize readInts24 performance for DocIdsWriter

["uschindler (via GitHub)" <gi...@apache.org>]

uschindler commented on issue #12527:
URL: https://github.com/apache/lucene/issues/12527#issuecomment-1708934703

   The main difference here is that by default Lucene uses MMapDirectoy. Loading many small arrays instead of single longs is not a good idea, as copyMemory off-heap to heap is heavy due to JVM overhead. When using MMapDirectory all the readLongs should be inlined, while those small long[] reads aren't.
   
   I wonder why you see higher syscalls in Opensearch, could it be because it uses hybridfs? Nowadays the use of hybridfs should be avoided as the old "number kernel mappings" limitations are no longer applicable since Java 19.
   
   So I disagree with chainging this loop of reading longs, it would be better to figure out why it is slower for you.


-- 
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-unsubscribe@lucene.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@lucene.apache.org
For additional commands, e-mail: issues-help@lucene.apache.org