You are viewing a plain text version of this content. The canonical link for it is here.
Posted to yarn-issues@hadoop.apache.org by "chaosju (Jira)" <ji...@apache.org> on 2021/05/24 12:05:00 UTC
[jira] [Resolved] (YARN-10784) Optimize RMNodeStatusEvent process
performance
[ https://issues.apache.org/jira/browse/YARN-10784?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
chaosju resolved YARN-10784.
----------------------------
Fix Version/s: 3.3.0
Resolution: Fixed
> Optimize RMNodeStatusEvent process performance
> ----------------------------------------------
>
> Key: YARN-10784
> URL: https://issues.apache.org/jira/browse/YARN-10784
> Project: Hadoop YARN
> Issue Type: Improvement
> Affects Versions: 2.8.5
> Reporter: chaosju
> Assignee: chaosju
> Priority: Major
> Fix For: 3.3.0
>
> Attachments: screenshot-2.png
>
>
> According to YARN-9927 analysis. The RMNodeStatusEvent dominates 90% time consumption of RM event scheduler.
> Use async-profiler profile the Processing of RMNodeStatusEvent,ProtoUtils.convertFromProtoFormat Use JDK String.replace().
> It is definitely worth considering not using these JDK String methods and opting for the faster Apache Commons StringUtils.replace() instead.
> Test case on jdk8:
> for (int i=0 ; i<100000 ;i++){
> ProtoUtils.convertFromProtoFormat(builder.getState());
> }
> Apache Commons StringUtils.replace() : used 30ms
> JDK String.replace() : used 60ms
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: yarn-issues-help@hadoop.apache.org