alamb commented on PR #20481:
URL: https://github.com/apache/datafusion/pull/20481#issuecomment-3958979921

   > I was reading the paper more in depth in recent days and concluded that 
the essence boils down to those things:
   
   I agree 
   
   > thread per core (in our case already true when setting target_partitions 
to cpu cores)
   
   Yes -- exactly -- and Tokio handles the scheudling
   
   > morsel = a large enough unit of work, large enough for communication 
overhead to be low, small enough to introduce enough parallelism
   
   In my mind, most of the plans use `RecordBatch` as the "Morsels" so a 
Stream<RecordBatch>` is very much a Morsel as described in the paper
   
   > work stealing / a queue with spreading "morsels" on the threads
   
   Indeed -- and in fact tokio has a [work stealing scheduler 
](https://tokio.rs/blog/2019-10-scheduler)
   
   In my mind, what this PR does is to break up Parquet files into smaller 
chunks that are read (but the granularity is bigger than a RecordBatch / target 
batch size, it is a RowGroup)


-- 
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