hadoop-mapreduce-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Arun C Murthy (Commented) (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (MAPREDUCE-3096) Add a good way to control the number of map/reduce tasks per node
Date Tue, 27 Sep 2011 07:33:12 GMT

    [ https://issues.apache.org/jira/browse/MAPREDUCE-3096?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13115287#comment-13115287

Arun C Murthy commented on MAPREDUCE-3096:

bq. As I said, please use the feature in the Scheduler - we don't want to support it per-Job,
has too many implications.

The main issue is that users can use this setting to hurt other tasks (of other jobs/users)
on the nodes. The CapacityScheduler prevents this by forcing the job to ask for more than
one slot per job, thus *charging* the job appropriately. Makes sense?
> Add a good way to control the number of map/reduce tasks per node
> -----------------------------------------------------------------
>                 Key: MAPREDUCE-3096
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-3096
>             Project: Hadoop Map/Reduce
>          Issue Type: Task
>            Reporter: Arsen Zahray
>             Fix For:
> Currently, controlling the number of map/reduce tasks is a hell.
> I've tried for it many times, and it doesn't work right. Also, I am not the only one
person, who seems to have this problem.
> There must be a better way to do it.
> Here's my proposal:
> add following functions to Job:
> setNumberOfMappersPerNode(int);
> setNumberOfReducersPerNode(int);
> setMaxMemoryPerMapper(int);
> setMaxMemoryPerReducer(int);

This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira


View raw message