[
https://issues.apache.org/jira/browse/SPARK-13131?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15131205#comment-15131205
]
Davies Liu commented on SPARK-13131:
------------------------------------
[~piccolbo] Thanks for you comments, we also have lots of offline discussion
and debates here. I'd like to provide more information here.
The benchmark, is a micro benchmark we used internal to measure some
performance improvements on CPU bound workload. Each benchmark will have
constant input (not randomly), ran after JVM is warmed up (the first run will
be dropped).
Because this micro benchmark is used as a tool in daily development, it should
not take too long to get a rough number, it also should not have too much noise
without many retires (or the daily development will be slow down much).
After some testing, We saw that the best time is much stable than average time,
also reasonable. So we (Reynold, Nong and me) decided to show both best time
and average time as result of benchmark.
> Use median time in benchmark
> ----------------------------
>
> Key: SPARK-13131
> URL: https://issues.apache.org/jira/browse/SPARK-13131
> Project: Spark
> Issue Type: Improvement
> Components: SQL
> Reporter: Davies Liu
> Assignee: Davies Liu
>
> Median time should be more stable than average time in benchmark.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
