[
https://issues.apache.org/jira/browse/ARROW-16716?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17545087#comment-17545087
]
Li Jin commented on ARROW-16716:
--------------------------------
[~ichauster] is an intern that is going to work on Acero benchmarks. I talked
to @Weston Pace offline and seems Projection is a good place to start.
Pasting [~westonpace] 's comments on this "
* Presumably, for a complex expression, with a large enough batch
size, the majority of time will be spent in the kernel functions.
* How far can you shrink the exec batch before the overhead of the
node impacts runtime?
* How complex does the expression need to be?
* I expect the results to be very similar to
ExecuteScalarExpressionOverhead in expression_benchmark.cc. Is it?
If not, what is the difference?
* What is the data rate of the project node (in bytes/second) for all
of the above?
* For all of the above run with both 1 thread and 1 thread per core
The ExecuteScalarExpressionOverhead benchmarks would be a good
existing example that should be pretty similar to how we benchmark
project node.
!https://ssl.gstatic.com/ui/v1/icons/mail/images/cleardot.gif!
"
> [Benchmarks] Create Projection benchmark for Acero
> --------------------------------------------------
>
> Key: ARROW-16716
> URL: https://issues.apache.org/jira/browse/ARROW-16716
> Project: Apache Arrow
> Issue Type: Improvement
> Reporter: Li Jin
> Priority: Major
>
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
