[
https://issues.apache.org/jira/browse/HADOOP-14407?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16014375#comment-16014375
]
Omkar Aradhya K S commented on HADOOP-14407:
--------------------------------------------
{quote}
Also, we will do more investigations into introducing both input and output
copybuffersize configurations.
{quote}
We did a small set of runs to see at what level the copybuffersize stagnates in
performance:
!TotalTime-vs-CopyBufferSize.jpg!
In this case, with copybuffersize set to just 128KB, we get >3x performance!
{quote}
If there is benefit of doing this I will submit a new patch with the changes or
else we will go ahead with this patch.
{quote}
Since even a small increase in the copybuffersize give the desired performance,
we will not need two separate copybuffersize for input and output.
> DistCp - Introduce a configurable copy buffer size
> --------------------------------------------------
>
> Key: HADOOP-14407
> URL: https://issues.apache.org/jira/browse/HADOOP-14407
> Project: Hadoop Common
> Issue Type: Improvement
> Components: tools/distcp
> Affects Versions: 2.9.0
> Reporter: Omkar Aradhya K S
> Assignee: Omkar Aradhya K S
> Fix For: 2.9.0, 3.0.0-alpha3
>
> Attachments: HADOOP-14407.001.patch, HADOOP-14407.002.patch,
> TotalTime-vs-CopyBufferSize.jpg
>
>
> Currently, the RetriableFileCopyCommand has a fixed copy buffer size of just
> 8KB. We have noticed in our performance tests that with bigger buffer sizes
> we saw upto ~3x performance boost. Hence, making the copy buffer size a
> configurable setting via the new parameter <copybuffersize>.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
---------------------------------------------------------------------
To unsubscribe, e-mail: common-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: common-issues-h...@hadoop.apache.org
