You are viewing a plain text version of this content. The canonical link for it is here.
Posted to common-dev@hadoop.apache.org by "Owen O'Malley (JIRA)" <ji...@apache.org> on 2006/05/07 05:55:20 UTC
[jira] Created: (HADOOP-200) The map task names are sent to the
reduces
The map task names are sent to the reduces
------------------------------------------
Key: HADOOP-200
URL: http://issues.apache.org/jira/browse/HADOOP-200
Project: Hadoop
Type: Bug
Components: mapred
Versions: 0.2
Reporter: Owen O'Malley
Assigned to: Owen O'Malley
Fix For: 0.3
As each reduce is created, it is given the entire set of potential map names. For my large sort jobs with 64k maps, this means that each reduce task is given a two dimensional array that is 5 tasks/map * 64k maps = 320k strings. Since the reduce task is passed from the job tracker to the task tracker and down to the task runner, passing the entire list is very expensive. I suspect that this is the cause of the slow downs that I see in the task trackers heart beats when the reduce tasks are being launched.
I propose that the ReduceTask be changed to just get the count of maps, with ids from 0 .. maps -1.
public ReduceTask(String jobFile, String taskId, int maps, int partition);
Then we need to change the protocol for finding map outputs:
MapOutputLocation[] locateMapOutputs(String jobId, int[] mapIds, int partition);
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see:
http://www.atlassian.com/software/jira
[jira] Commented: (HADOOP-200) The map task names are sent to the
reduces
Posted by "Doug Cutting (JIRA)" <ji...@apache.org>.
[ http://issues.apache.org/jira/browse/HADOOP-200?page=comments#action_12378484 ]
Doug Cutting commented on HADOOP-200:
-------------------------------------
+1 this sounds good to me.
> The map task names are sent to the reduces
> ------------------------------------------
>
> Key: HADOOP-200
> URL: http://issues.apache.org/jira/browse/HADOOP-200
> Project: Hadoop
> Type: Bug
> Components: mapred
> Versions: 0.2
> Reporter: Owen O'Malley
> Assignee: Owen O'Malley
> Fix For: 0.3
>
> As each reduce is created, it is given the entire set of potential map names. For my large sort jobs with 64k maps, this means that each reduce task is given a two dimensional array that is 5 tasks/map * 64k maps = 320k strings. Since the reduce task is passed from the job tracker to the task tracker and down to the task runner, passing the entire list is very expensive. I suspect that this is the cause of the slow downs that I see in the task trackers heart beats when the reduce tasks are being launched.
> I propose that the ReduceTask be changed to just get the count of maps, with ids from 0 .. maps -1.
> public ReduceTask(String jobFile, String taskId, int maps, int partition);
> Then we need to change the protocol for finding map outputs:
> MapOutputLocation[] locateMapOutputs(String jobId, int[] mapIds, int partition);
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see:
http://www.atlassian.com/software/jira
[jira] Commented: (HADOOP-200) The map task names are sent to the
reduces
Posted by "eric baldeschwieler (JIRA)" <ji...@apache.org>.
[ http://issues.apache.org/jira/browse/HADOOP-200?page=comments#action_12378309 ]
eric baldeschwieler commented on HADOOP-200:
--------------------------------------------
+1 and another and another...
> The map task names are sent to the reduces
> ------------------------------------------
>
> Key: HADOOP-200
> URL: http://issues.apache.org/jira/browse/HADOOP-200
> Project: Hadoop
> Type: Bug
> Components: mapred
> Versions: 0.2
> Reporter: Owen O'Malley
> Assignee: Owen O'Malley
> Fix For: 0.3
>
> As each reduce is created, it is given the entire set of potential map names. For my large sort jobs with 64k maps, this means that each reduce task is given a two dimensional array that is 5 tasks/map * 64k maps = 320k strings. Since the reduce task is passed from the job tracker to the task tracker and down to the task runner, passing the entire list is very expensive. I suspect that this is the cause of the slow downs that I see in the task trackers heart beats when the reduce tasks are being launched.
> I propose that the ReduceTask be changed to just get the count of maps, with ids from 0 .. maps -1.
> public ReduceTask(String jobFile, String taskId, int maps, int partition);
> Then we need to change the protocol for finding map outputs:
> MapOutputLocation[] locateMapOutputs(String jobId, int[] mapIds, int partition);
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see:
http://www.atlassian.com/software/jira
[jira] Closed: (HADOOP-200) The map task names are sent to the
reduces
Posted by "Doug Cutting (JIRA)" <ji...@apache.org>.
[ http://issues.apache.org/jira/browse/HADOOP-200?page=all ]
Doug Cutting closed HADOOP-200:
-------------------------------
> The map task names are sent to the reduces
> ------------------------------------------
>
> Key: HADOOP-200
> URL: http://issues.apache.org/jira/browse/HADOOP-200
> Project: Hadoop
> Type: Bug
> Components: mapred
> Versions: 0.2
> Reporter: Owen O'Malley
> Assignee: Owen O'Malley
> Fix For: 0.3.0
> Attachments: map-id.patch
>
> As each reduce is created, it is given the entire set of potential map names. For my large sort jobs with 64k maps, this means that each reduce task is given a two dimensional array that is 5 tasks/map * 64k maps = 320k strings. Since the reduce task is passed from the job tracker to the task tracker and down to the task runner, passing the entire list is very expensive. I suspect that this is the cause of the slow downs that I see in the task trackers heart beats when the reduce tasks are being launched.
> I propose that the ReduceTask be changed to just get the count of maps, with ids from 0 .. maps -1.
> public ReduceTask(String jobFile, String taskId, int maps, int partition);
> Then we need to change the protocol for finding map outputs:
> MapOutputLocation[] locateMapOutputs(String jobId, int[] mapIds, int partition);
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see:
http://www.atlassian.com/software/jira
[jira] Updated: (HADOOP-200) The map task names are sent to the
reduces
Posted by "Owen O'Malley (JIRA)" <ji...@apache.org>.
[ http://issues.apache.org/jira/browse/HADOOP-200?page=all ]
Owen O'Malley updated HADOOP-200:
---------------------------------
Attachment: map-id.patch
Here is the patch. It does dramatically lower the slugishness when the reduces are launching.
> The map task names are sent to the reduces
> ------------------------------------------
>
> Key: HADOOP-200
> URL: http://issues.apache.org/jira/browse/HADOOP-200
> Project: Hadoop
> Type: Bug
> Components: mapred
> Versions: 0.2
> Reporter: Owen O'Malley
> Assignee: Owen O'Malley
> Fix For: 0.3
> Attachments: map-id.patch
>
> As each reduce is created, it is given the entire set of potential map names. For my large sort jobs with 64k maps, this means that each reduce task is given a two dimensional array that is 5 tasks/map * 64k maps = 320k strings. Since the reduce task is passed from the job tracker to the task tracker and down to the task runner, passing the entire list is very expensive. I suspect that this is the cause of the slow downs that I see in the task trackers heart beats when the reduce tasks are being launched.
> I propose that the ReduceTask be changed to just get the count of maps, with ids from 0 .. maps -1.
> public ReduceTask(String jobFile, String taskId, int maps, int partition);
> Then we need to change the protocol for finding map outputs:
> MapOutputLocation[] locateMapOutputs(String jobId, int[] mapIds, int partition);
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see:
http://www.atlassian.com/software/jira
[jira] Resolved: (HADOOP-200) The map task names are sent to the
reduces
Posted by "Doug Cutting (JIRA)" <ji...@apache.org>.
[ http://issues.apache.org/jira/browse/HADOOP-200?page=all ]
Doug Cutting resolved HADOOP-200:
---------------------------------
Resolution: Fixed
I just committed this. Thanks, Owen!
> The map task names are sent to the reduces
> ------------------------------------------
>
> Key: HADOOP-200
> URL: http://issues.apache.org/jira/browse/HADOOP-200
> Project: Hadoop
> Type: Bug
> Components: mapred
> Versions: 0.2
> Reporter: Owen O'Malley
> Assignee: Owen O'Malley
> Fix For: 0.3
> Attachments: map-id.patch
>
> As each reduce is created, it is given the entire set of potential map names. For my large sort jobs with 64k maps, this means that each reduce task is given a two dimensional array that is 5 tasks/map * 64k maps = 320k strings. Since the reduce task is passed from the job tracker to the task tracker and down to the task runner, passing the entire list is very expensive. I suspect that this is the cause of the slow downs that I see in the task trackers heart beats when the reduce tasks are being launched.
> I propose that the ReduceTask be changed to just get the count of maps, with ids from 0 .. maps -1.
> public ReduceTask(String jobFile, String taskId, int maps, int partition);
> Then we need to change the protocol for finding map outputs:
> MapOutputLocation[] locateMapOutputs(String jobId, int[] mapIds, int partition);
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see:
http://www.atlassian.com/software/jira