hadoop-common-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Arun C Murthy (JIRA)" <j...@apache.org>
Subject [jira] Updated: (HADOOP-1274) Configuring different number of mappers and reducers per TaskTracker
Date Fri, 26 Oct 2007 11:31:48 GMT

     [ https://issues.apache.org/jira/browse/HADOOP-1274?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Arun C Murthy updated HADOOP-1274:
----------------------------------

    Fix Version/s: 0.16.0
           Status: Open  (was: Patch Available)

Comments:

1. Conversion to short here is unnecessary, probably dangerous:
{noformat}
-          metricsRecord.setMetric("taskSlots", (short)maxCurrentTasks);
+          metricsRecord.setMetric("mapTaskSlots", (short)maxCurrentMapTasks);
+          metricsRecord.setMetric("reduceTaskSlots", 
+                                      (short)maxCurrentReduceTasks);
{noformat}

2. Please deprecate {{mapred.tasktracker.tasks.maximum}} for this release, as per existing
norms. We cannot remove it right-away. {{mapred.tasktracker.tasks.maximum}} should be superceded
by {{mapred.tasktracker.map.tasks.maximum} and {{mapred.tasktracker.reduce.tasks.maximum}}
for 0.16.0.

{noformat}
-    maxCurrentTasks = conf.getInt("mapred.tasktracker.tasks.maximum", 2);
+    maxCurrentMapTasks = conf.getInt(
+                             "mapred.tasktracker.map.tasks.maximum", 2);
+    maxCurrentReduceTasks = conf.getInt(
+                             "mapred.tasktracker.reduce.tasks.maximum", 1);
{noformat}

3. Should we consider having different {{JobTracker.PAD_FRACTION}} for maps and reduces? Clearly
the no. of padded slots for reduces should be higher since reduce-failures are more expensive...

> Configuring different number of mappers and reducers per TaskTracker
> --------------------------------------------------------------------
>
>                 Key: HADOOP-1274
>                 URL: https://issues.apache.org/jira/browse/HADOOP-1274
>             Project: Hadoop
>          Issue Type: Improvement
>          Components: mapred
>            Reporter: Koji Noguchi
>            Assignee: Amareshwari Sri Ramadasu
>            Priority: Minor
>             Fix For: 0.16.0
>
>         Attachments: patch-1274.txt
>
>
> Depending on the application, it sometimes make sense to have more mappers than reducers
assigned to each node. 
> (I'm assuming user either has a dedicated cluster or use HOD.)

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


Mime
View raw message