[
https://issues.apache.org/jira/browse/HDDS-7228?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17769186#comment-17769186
]
Kirill Sizov commented on HDDS-7228:
------------------------------------
Hi [~weichiu],
I came across the same issue during perf testing of Ozone.
I fixed it locally and here are the results.
before the fix
!before.png!
And after the fix
!after.png!
The fix not only reduced the CPU usage, but also improved total throughput.
Before the fix the increase of clients from 75 to 150 per node led to the
decline of RPS: from 480 files per sec to 410 files per sec.
After the fix I see the opposite - the number of files per second increased
form 480 to 580.
> ChecksumByteBufferImpl.update() is expensive
> --------------------------------------------
>
> Key: HDDS-7228
> URL: https://issues.apache.org/jira/browse/HDDS-7228
> Project: Apache Ozone
> Issue Type: Improvement
> Components: Ozone Client
> Affects Versions: 1.3.0
> Reporter: Wei-Chiu Chuang
> Assignee: Wei-Chiu Chuang
> Priority: Major
> Labels: pull-request-available
> Attachments: Screen Shot 2022-08-22 at 7.04.42 PM.png, after.png,
> before.png
>
>
> ChecksumByteBufferImpl.update() involves memory copy.
> It looks like the memory copy is redundant, and checksum should be able to
> compute directly.
> I have a hack to force it to skip the memory copy so I'll post it as a demo.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
