[ 
https://issues.apache.org/jira/browse/MAPREDUCE-2876?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13094562#comment-13094562
 ] 

Robert Joseph Evans commented on MAPREDUCE-2876:
------------------------------------------------

I am fine with the polling like I said before.  The question I was trying to 
answer was what is the ramifications of not having a polling interval, but just 
having an expiry interval, and setting the polling interval off of the expiry 
interval.  It means the real expiry interval is any time between expiryInterval 
and expiryInterval + pollingInterval.  If that is never a problem then polling 
is fine. 

I can see your reasoning behind the change.  When there are N objects being 
tracked then polling would be

{code}
 O((pollingFrequency * N) + (modificationFrequency * N))
{code}

But Priority Queue would be

{code}
O(modificationFrequency * N * log(N) + expiryFrequency)
{code}

So if the modificationFrequency is very large then polling definitely wins in 
complexity, but I am curious did you run any benchmarks to see what kind of an 
extra load would be placed on the system with a priority queue (Just curious 
what it might be)?

> ContainerAllocationExpirer appears to use the incorrect configs
> ---------------------------------------------------------------
>
>                 Key: MAPREDUCE-2876
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-2876
>             Project: Hadoop Map/Reduce
>          Issue Type: Bug
>          Components: mrv2
>    Affects Versions: 0.23.0
>            Reporter: Robert Joseph Evans
>            Assignee: Robert Joseph Evans
>             Fix For: 0.23.0
>
>
> ContainerAllocationExpirer sets the expiration interval to be 
> RMConfig.CONTAINER_LIVELINESS_MONITORING_INTERVAL but uses 
> AMLIVELINESS_MONITORING_INTERVAL as the interval.  This is very different 
> from what AMLivelinessMonitor does.
> There should be two configs RMConfig.CONTAINER_LIVELINESS_MONITORING_INTERVAL 
> for the monitoring interval and RMConfig.CONTAINER_EXPIRY_INTERVAL for the 
> expiry.

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira

        

Reply via email to