[
https://issues.apache.org/jira/browse/HADOOP-2141?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12542867
]
Arun C Murthy commented on HADOOP-2141:
---------------------------------------
bq. Is there a strong reason for disabling it for maps ?
To me it looks like there is strong case for atleast treating maps and reduces
separately, if not disabling for maps.
Mostly maps run in the order of minutes, while reduces take much more time
since they are waiting for maps to complete.
E.g In sort500, maps take about a minute (outlier maps take 4mins) while
reduces (atleast the first wave) take around an hour to complete.
Given these (and I know there are wildly varying job characteristics) I'd like
to be careful and ensure we aren't too aggressive while launching speculative,
speculative-tasks (man! does that sound weird!).
Hence, and keeping in mind that reduces are more expensive to execute
willy-nilly speculatively I propose 4 parameters, to keep it reasonably simple:
mapred.map.speculative.timegap = 2 x avg_map_completion_time
mapred.reduce.speculative.timegap = 1.5 x avg_reduce_completion_time
mapred.map.min.completion.for.speculation = 90
mapred.reduce.min.completion.for.speculation = 95
Thoughts?
> speculative execution start up condition based on completion time
> -----------------------------------------------------------------
>
> Key: HADOOP-2141
> URL: https://issues.apache.org/jira/browse/HADOOP-2141
> Project: Hadoop
> Issue Type: Improvement
> Components: mapred
> Affects Versions: 0.15.0
> Reporter: Koji Noguchi
> Assignee: Arun C Murthy
> Fix For: 0.16.0
>
>
> We had one job with speculative execution hang.
> 4 reduce tasks were stuck with 95% completion because of a bad disk.
> Devaraj pointed out
> bq . One of the conditions that must be met for launching a speculative
> instance of a task is that it must be at least 20% behind the average
> progress, and this is not true here.
> It would be nice if speculative execution also starts up when tasks stop
> making progress.
> Devaraj suggested
> bq. Maybe, we should introduce a condition for average completion time for
> tasks in the speculative execution check.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
