azagrebin commented on pull request #13628:
URL: https://github.com/apache/flink/pull/13628#issuecomment-709881257


   > 2. v1 and v3 do not have input dependency, so putting them into the same 
slot does not reduce data exchange cost, compared to tasks in the same 
pipelined region
   
   I meant execution speed up by running V1 and v3 at the same time in the same 
slot (makes sense w/o dynamic slot allocation as it is mostly the case now).
   
   > 3. If we can have one slot which runs v11 and v31 together, it indicates 
that the resources are actually enough to be split into 2 slots, one for v11 
and one for v31.
   
   This is indeed not an issue when we have dynamic slot allocation.
   
   > So I think assigning each logical pipelined region to a different 
SlotSharingGroup is enough and simpler.
   > Moreover, as you mentioned, this is just the default behavior and we can 
enable users to set `SlotSharingGroup` for tasks via API if there were strong 
requirements for that.
   
   Agree, it is generally hard for Flink to decide which regions can run 
together in batch.


----------------------------------------------------------------
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.

For queries about this service, please contact Infrastructure at:
[email protected]


Reply via email to