costin commented on PR #15208:
URL: https://github.com/apache/lucene/pull/15208#issuecomment-5625017392

   Hi folks,
   
   Fyi, I've been working on a different approach to avoid the expensive full 
beam computation on the concurrent merge path at #16664. 
   Instead of computing eps on the main thread at enqueue time, workers share 
completion state through a bitset (cheap) and use it when computing their own 
eps. Each worker checks which source-graph neighbors are already connected in 
the output graph and uses their neighborhoods as entry points.
   
   It's an opportunistic approach that seems to work well as the workers start 
converging (similar to the global competitive query score). It avoids the 
staleness problem identified in this thread. The v < u ordering constraint is 
replaced by the completion check, so no ordering assumptions or serialization 
are needed across workers.
   
   Feedback welcome!


-- 
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]

Reply via email to