[
https://issues.apache.org/jira/browse/YARN-1011?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13726965#comment-13726965
]
Bikas Saha commented on YARN-1011:
----------------------------------
An approach that has been successfully used elsewhere is to allocate all
guaranteed work based on expected resource utilization (eg from cluster
statistics) and then measure the actual usage. Schedule non-guaranteed work at
lower priority to soak up the excess capacity. The lower priority ensures that
the non-guaranteed work minimally interferes with the guaranteed work when the
guaranteed work requires resources. In some sense this can be called
over-subscription. Non fluid resources like memory tend to be ones that
typically cause preemption of non-guaranteed work when the over-subscription
may result in performance loss.
> [Umbrella] RM should dynamically schedule containers based on utilization of
> currently allocated containers
> -----------------------------------------------------------------------------------------------------------
>
> Key: YARN-1011
> URL: https://issues.apache.org/jira/browse/YARN-1011
> Project: Hadoop YARN
> Issue Type: New Feature
> Reporter: Arun C Murthy
> Assignee: Arun C Murthy
>
> Currently RM allocates containers and assumes resources allocated are
> utilized.
> RM can, and should, get to a point where it measures utilization of allocated
> containers and, if appropriate, allocate more (speculative?) containers.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
