[
https://issues.apache.org/jira/browse/MESOS-3137?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14640738#comment-14640738
]
Benjamin Mahler commented on MESOS-3137:
----------------------------------------
(linked in task resizing)
Are you running your own scheduler? Can you model the "startup" work as an
additional task on top that finishes once the regular task reaches steady state?
> Variable CPU allocation for task startup vs steady state
> --------------------------------------------------------
>
> Key: MESOS-3137
> URL: https://issues.apache.org/jira/browse/MESOS-3137
> Project: Mesos
> Issue Type: Improvement
> Reporter: Barak Shechter
>
> We are currently experiencing long startup times because we are requesting
> CPU based on what tasks' during the normal course of operation.
> As a workaround, we have given our tasks a longer grace period and more CPU
> so that they start up in a more reasonable amount of time and are not
> terminated as quickly. Unfortunately, this leads to underutilization of
> resources because our applications don't require the higher cpu fraction
> after they've bootstrapped.
> We would like to be able to specify a higher CPU allotment for task startup
> which can then be reduced for steady state operation of the task.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
