[ 
https://issues.apache.org/jira/browse/YARN-4995?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15426123#comment-15426123
 ] 

stefanlee commented on YARN-4995:
---------------------------------

[~xupeng] thanks your jira, i have a question,the demand resource you mentioned 
will beyond the max resource a lot,IMO,the demand resource is used resource 
plus request resource,because of node locality and rack locality ,a task's 
request resource may include many containers,e.g. 
<20,"node1","memory:1G,cpu:1",1,true> ,<20,"node2","memory:1G,cpu:1",1,true> 
,<20,"rack1","memory:1G,cpu:1",1,true> ,<20,"rack2","memory:1G,cpu:1",1,true> 
,<20,"*","memory:1G,cpu:1",1,true> , a MR application may has 20 maps and  need 
20*<memory: 1G,cpu:1>,but the demand resource will reach 5*20*<memory:1G,cpu:1> 
,so the demand resource cant help user to modify  a queue's min/max resource 
clearly. I dont know my option correct or not?

> FairScheduler: Display per-queue demand on the scheduler page
> -------------------------------------------------------------
>
>                 Key: YARN-4995
>                 URL: https://issues.apache.org/jira/browse/YARN-4995
>             Project: Hadoop YARN
>          Issue Type: Improvement
>            Reporter: xupeng
>            Assignee: xupeng
>            Priority: Minor
>             Fix For: 2.8.0
>
>         Attachments: YARN-4995.001.patch, YARN-4995.002.patch, 
> demo_screenshot.png
>
>
> For now there is no demand resource information for queues on the scheduler 
> page. 
> Just using used resource information, it is hard for us to judge whether the 
> queue is needy (demand > used , but cluster has no available resource). And 
> without demand resource information, modifying min/max resource for queue is 
> not accurate. 



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org

Reply via email to