Re: [I] perf: hybrid sort-shuffle buffering: split rows by partition on arrival when memory allows [datafusion-ballista]

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

NoahKusaba commented on issue #2578:
URL: https://github.com/apache/datafusion-ballista/issues/2578#issuecomment-6093596490

   Hey @andygrove I know you're busy, but do you have time to eyeball / smell-test this proposal? I know you have prior experience investigating this exact solution on Comet (see above), and would appreciate your input.
   
   I've spent alot of time thinking about this solution. The TPCH - SF10 benchmark seems to indicate a 8.5 -> 10% increase in performance, and I think this hybrid approach addresses scalability concerns you previously ran into (copying Gluten). Maybe this will be easier if I just opened a PR, so you have the code to read through?
   
   I'll split out two other PR's while working on the branch, which includes an elegant refactor / minor perf improvement + shuffle statistic fix that can land separately regardless of the resolution. 


-- 
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: github-unsubscribe@datafusion.apache.org

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


---------------------------------------------------------------------
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For additional commands, e-mail: github-help@datafusion.apache.org