[
https://issues.apache.org/jira/browse/MESOS-2652?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14627096#comment-14627096
]
Christos Kozyrakis commented on MESOS-2652:
-------------------------------------------
You are correct Jie, the lower you set the shares for the BE tasks, the better
it will be. If you give the BE group 1:1024, the BE tasks will get 0.1% of the
CPU time over the long term (assuming other tasks can consume the rest 99.9%).
This is a perfectly good solution to begin with, The high priority throughput
tasks will never even notice. Some low latency tasks will but it's not a bad
starting point at all.
However, every now and then you will see some glitches on low latency tasks.
Even if the HP tasks are 100% busy, the 1:1024 setting will allow the BE task
to run eventually, introducing a glitch of a few msec. Keep this in mind. If at
some point this becomes an issue, there are several ways to deal with this:
- disallow oversubscription for the (hopefully small % of) low latency apps
that care about it
- fix SCHED_IDLE somehow
- use CPU_sets
- ...
> Update Mesos containerizer to understand revocable cpu resources
> ----------------------------------------------------------------
>
> Key: MESOS-2652
> URL: https://issues.apache.org/jira/browse/MESOS-2652
> Project: Mesos
> Issue Type: Task
> Reporter: Vinod Kone
> Assignee: Ian Downes
> Labels: twitter
> Fix For: 0.23.0
>
> Attachments: Abnormal performance with 3 additional revocable tasks
> (1).png, Abnormal performance with 3 additional revocable tasks (2).png,
> Abnormal performance with 3 additional revocable tasks (3).png, Abnormal
> performance with 3 additional revocable tasks (4).png, Abnormal performance
> with 3 additional revocable tasks (5).png, Abnormal performance with 3
> additional revocable tasks (6).png, Abnormal performance with 3 additional
> revocable tasks (7).png, Performance improvement after reducing cpu.share to
> 2 for revocable tasks (1).png, Performance improvement after reducing
> cpu.share to 2 for revocable tasks (10).png, Performance improvement after
> reducing cpu.share to 2 for revocable tasks (2).png, Performance improvement
> after reducing cpu.share to 2 for revocable tasks (3).png, Performance
> improvement after reducing cpu.share to 2 for revocable tasks (4).png,
> Performance improvement after reducing cpu.share to 2 for revocable tasks
> (5).png, Performance improvement after reducing cpu.share to 2 for revocable
> tasks (6).png, Performance improvement after reducing cpu.share to 2 for
> revocable tasks (7).png, Performance improvement after reducing cpu.share to
> 2 for revocable tasks (8).png, Performance improvement after reducing
> cpu.share to 2 for revocable tasks (9).png
>
>
> The CPU isolator needs to properly set limits for revocable and non-revocable
> containers.
> The proposed strategy is to use a two-way split of the cpu cgroup hierarchy
> -- normal (non-revocable) and low priority (revocable) subtrees -- and to use
> a biased split of CFS cpu.shares across the subtrees, e.g., a 20:1 split
> (TBD). Containers would be present in only one of the subtrees. CFS quotas
> will *not* be set on subtree roots, only cpu.shares. Each container would set
> CFS quota and shares as done currently.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
