You are viewing a plain text version of this content. The canonical link for it is here.

Posted to issues@mesos.apache.org by "Benjamin Mahler (JIRA)" <ji...@apache.org> on 2017/02/15 22:33:41 UTC

[jira] [Updated] (MESOS-7132) Fair share can be exceeded due to coarse-grained offer technique.

     [ https://issues.apache.org/jira/browse/MESOS-7132?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Benjamin Mahler updated MESOS-7132:
-----------------------------------
       Priority: Major  (was: Critical)
    Description: 
The current implementation of fair share allocation allocates the entire available resources on an agent when trying to satisfy the fair share. What this means is that the fair share can be exceeded by the size of an agent.

This is especially problematic for large machines, consider a 48 core, 512 GB memory server where a role's share is 4 cores and 4GB of memory. Given our current approach, we will send an offer for the entire 48 cores and 512 GB of memory!

This ticket is to perform fine grained offers when the allocation will exceed the fair share.

  was:
The current implementation of quota allocation allocates the entire available resources on an agent when trying to satisfy the quota. What this means is that quota can be exceeded by the size of an agent.

This is especially problematic for large machines, consider a 48 core, 512 GB memory server where a role is given 4 cores and 4GB of memory. Given our current approach, we will send an offer for the entire 48 cores and 512 GB of memory!

This ticket is to perform fine grained offers when the allocation will exceed the quota.


> Fair share can be exceeded due to coarse-grained offer technique.
> -----------------------------------------------------------------
>
>                 Key: MESOS-7132
>                 URL: https://issues.apache.org/jira/browse/MESOS-7132
>             Project: Mesos
>          Issue Type: Bug
>          Components: allocation
>            Reporter: Benjamin Mahler
>
> The current implementation of fair share allocation allocates the entire available resources on an agent when trying to satisfy the fair share. What this means is that the fair share can be exceeded by the size of an agent.
> This is especially problematic for large machines, consider a 48 core, 512 GB memory server where a role's share is 4 cores and 4GB of memory. Given our current approach, we will send an offer for the entire 48 cores and 512 GB of memory!
> This ticket is to perform fine grained offers when the allocation will exceed the fair share.



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)