[jira] Commented: (HADOOP-4001) TaskTracker's Memory resource should be considered when tasktracker asks for new task

["ZhuGuanyin (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HADOOP-4001?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12624647#action_12624647 ] 

ZhuGuanyin commented on HADOOP-4001:
------------------------------------

    Currently, taskTracker only considers enough free disk space left when it asks for new task, memory resource should be considered too, or it may works badly in SMP environment. 

    Let's consider a scenario: one smp tasktracker with 8 cpu and 8 GB memory, and mapred.tasktracker.tasks.maximum=7. It has running 5 tasks while each task consumed 1.5GB memory, so it would asks for 2 new tasks which would make things go badly. 

    We should add free memory check just like 
         askForNewTask = enoughFreeMem(localMinMemStart);   in tasktracker.java just like 
         askForNewTask = enoughFreeSpace(localMinSpaceStart);

     I will attach a patch for this feature as soon as possible.



> TaskTracker's Memory resource should be considered when tasktracker asks for new task
> -------------------------------------------------------------------------------------
>
>                 Key: HADOOP-4001
>                 URL: https://issues.apache.org/jira/browse/HADOOP-4001
>             Project: Hadoop Core
>          Issue Type: New Feature
>          Components: mapred
>    Affects Versions: 0.17.2
>            Reporter: ZhuGuanyin
>            Priority: Minor
>             Fix For: 0.19.0
>
>
> Currently, taskTracker only considers enough free disk space left when it asks for new task, memory resource should be considered too, or it may works badly in SMP environment.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.