You are viewing a plain text version of this content. The canonical link for it is here.

Posted to mapreduce-issues@hadoop.apache.org by "Allen Wittenauer (JIRA)" <ji...@apache.org> on 2014/07/22 21:16:40 UTC

[jira] [Commented] (MAPREDUCE-327) Add explicit remote map count JobTracker metrics

    [ https://issues.apache.org/jira/browse/MAPREDUCE-327?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14070734#comment-14070734 ] 

Allen Wittenauer commented on MAPREDUCE-327:
--------------------------------------------

In real-world scenarios, we've discovered that task locality as reported by the system can effectively be a lie because of CFIF/MFIF. Given 4 input splits, if the first is local but the rest are not, the task will still be considered local even though 3/4'ths of the data came off rack!

> Add explicit remote map count JobTracker metrics
> ------------------------------------------------
>
>                 Key: MAPREDUCE-327
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-327
>             Project: Hadoop Map/Reduce
>          Issue Type: Improvement
>            Reporter: Hong Tang
>              Labels: newbie
>
> I am proposing to add a counter REMOTE_MAPS in addition to the following counters: TOTAL_MAPS, DATA_LOCAL_MAPS, RACK_LOCAL_MAPS. A Map Task is considered a "remote-map" iff the input split returns a set of locations, but none is chosen to execute the map task.



--
This message was sent by Atlassian JIRA
(v6.2#6252)