benwtrent commented on PR #14167:
URL: https://github.com/apache/lucene/pull/14167#issuecomment-2616760637

   > This made me wonder if it would be a better trade-off to let just one 
slice run on its own first, and then let all other N-1 slices run in parallel 
with one another,
   
   I really like this idea. For kNN search, it seems best to take the largest 
tiers, gather information from them, and then run the smaller tiers in 
parallel. 
   
   The major downside of kNN is that there is no slicing at all. Every segment 
is just its own worker, which is sort of crazy. We should at a minimum combine 
all the tiny segments together into a single thread.
   
   What do you think @mayya-sharipova ? Slicing the segments and then picking 
the "largest" slice and search that in current thread. Then using that 
information to help the future parallel threads? 


-- 
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

Reply via email to