[
https://issues.apache.org/jira/browse/FLINK-23826?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17414833#comment-17414833
]
Zhu Zhu commented on FLINK-23826:
---------------------------------
Here's the testing result:
||JM/Scheduler Initializing||Task deployment||Making failure recovery decision||
|0.2s|16s|0.2s|
A similar benchmark against Flink-1.12(which does not include all the
performance improvment) is in progress but blocked by some environment
problems. It will be updated a bit later.
And this ticket is no longer a blocker of Flink-1.14.
> Verify optimized scheduler performance for large-scale jobs
> -----------------------------------------------------------
>
> Key: FLINK-23826
> URL: https://issues.apache.org/jira/browse/FLINK-23826
> Project: Flink
> Issue Type: Sub-task
> Components: Runtime / Coordination
> Affects Versions: 1.14.0
> Reporter: Zhu Zhu
> Assignee: Zhu Zhu
> Priority: Critical
> Fix For: 1.14.0
>
>
> This ticket is used to verify the result of FLINK-21110.
> It should check if large scale jobs' scheduling are working well and the
> scheduling performance, with a real job running on cluster.
> The conclusion should include, for a *10000 --- all-to-all-connected
> -->10000* job:
> 1. time of job initialization on master (job received -> scheduling started)
> 2. time of task deployment (task deploying started -> all tasks in
> INITIALIZATION)
> 3. time of making task failure recovery decision (JM notified about task
> failure -> tasks to restart decided)
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
