[ 
https://issues.apache.org/jira/browse/HADOOP-1338?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12661665#action_12661665
 ] 

Matei Zaharia commented on HADOOP-1338:
---------------------------------------

I think explicitly fetching multiple maps per connection is the best way to go. 
We can control exactly what happens instead of relying on connection keep-alive 
to work in the HTTP library and running into problems if this changes. In terms 
of how many outputs to fetch, I believe we should take the same approach as 
Dhruba's multiple-splits-per-map patch - don't say a fixed number of outputs, 
but rather a fixed number of bytes (say up to 256 MB). If a job has small map 
outputs, you should fetch more, and in the extreme, for something like 
webdatascan, you might as well fetch all the outputs on a host in one go.

> Improve the shuffle phase by using the "connection: keep-alive" and doing 
> batch transfers of files
> --------------------------------------------------------------------------------------------------
>
>                 Key: HADOOP-1338
>                 URL: https://issues.apache.org/jira/browse/HADOOP-1338
>             Project: Hadoop Core
>          Issue Type: Improvement
>          Components: mapred
>            Reporter: Devaraj Das
>            Assignee: Jothi Padmanabhan
>
> We should do transfers of map outputs at the granularity of  
> *total-bytes-transferred* rather than the current way of transferring a 
> single file and then closing the connection to the server. A single 
> TaskTracker might have a couple of map output files for a given reduce, and 
> we should transfer multiple of them (upto a certain total size) in a single 
> connection to the TaskTracker. Using HTTP-1.1's keep-alive connection would 
> help since it would keep the connection open for more than one file transfer. 
> We should limit the transfers to a certain size so that we don't hold up a 
> jetty thread indefinitely (and cause timeouts for other clients).
> Overall, this should give us improved performance.

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

Reply via email to