[ 
https://issues.apache.org/jira/browse/HADOOP-1336?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Marco Nicosia updated HADOOP-1336:
----------------------------------


At the risk of stirring up closed issues, are we sure this is the correct 
choice? We've been keeping speculative execution off on our clusters, and 
recently accidentally turned it on. Drama ensued. The way I see it, there are 
two scenarios for users who don't know about speculative execution. Either it's 
off by default, and it's no problem to anyone, or it's on by default, and a 
subset are bewildered by Hadoop's "odd" behavior. This is a support burden.

The only reason to have it on by default (I think?) is to improve performance 
across the board. But this is by far not the only solution to affect 
performance, so we are doing significant user auditing regardless. If our 
less-efficient users can take advantage of speculative execution for their 
jobs, they will. I think, since some knowledge of speculative execution is good 
before enabling it, it's better used as part of the optimization process, 
rather than making it a default of the framework.


> turn on speculative execution by defaul
> ---------------------------------------
>
>                 Key: HADOOP-1336
>                 URL: https://issues.apache.org/jira/browse/HADOOP-1336
>             Project: Hadoop
>          Issue Type: Task
>          Components: mapred
>    Affects Versions: 0.12.3
>            Reporter: Owen O'Malley
>            Assignee: Owen O'Malley
>             Fix For: 0.14.0
>
>         Attachments: spec-exec.patch
>
>
> Now that speculative execution is working again, we should enable it by 
> default.

-- 
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