[jira] [Commented] (MESOS-314) Support the cgroups 'cpusets' subsystem.

2016-08-16 Thread Charles Allen (JIRA)

[ 
https://issues.apache.org/jira/browse/MESOS-314?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15423043#comment-15423043
 ] 

Charles Allen commented on MESOS-314:
-

Needs more experimentation / validation though.

> Support the cgroups 'cpusets' subsystem.
> 
>
> Key: MESOS-314
> URL: https://issues.apache.org/jira/browse/MESOS-314
> Project: Mesos
>  Issue Type: Story
>Reporter: Benjamin Mahler
>  Labels: twitter
>
> We'd like to add support for the cpusets subsystem, in order to support 
> pinning to cpus.
> This has several potential benefits:
> 1. Improved isolation against other tenants, when given exclusive access to 
> cores.
> 2. Improved performance, if pinned to several cores with good locality in the 
> CPU topology.
> 3. An alternative / complement to CFS for applying an upper limit on CPU 
> usage.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)


[jira] [Commented] (MESOS-314) Support the cgroups 'cpusets' subsystem.

2016-08-16 Thread Charles Allen (JIRA)

[ 
https://issues.apache.org/jira/browse/MESOS-314?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15423040#comment-15423040
 ] 

Charles Allen commented on MESOS-314:
-

Druid.io is having big-node scaling issues reported by the community (see 
https://groups.google.com/d/msg/druid-user/l5grbWf2x9w/nF5fPyiaBQAJ ) I think 
having proper cpuset / memory controller isolation per task would help with 
this tremendously. 

> Support the cgroups 'cpusets' subsystem.
> 
>
> Key: MESOS-314
> URL: https://issues.apache.org/jira/browse/MESOS-314
> Project: Mesos
>  Issue Type: Story
>Reporter: Benjamin Mahler
>  Labels: twitter
>
> We'd like to add support for the cpusets subsystem, in order to support 
> pinning to cpus.
> This has several potential benefits:
> 1. Improved isolation against other tenants, when given exclusive access to 
> cores.
> 2. Improved performance, if pinned to several cores with good locality in the 
> CPU topology.
> 3. An alternative / complement to CFS for applying an upper limit on CPU 
> usage.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)


[jira] [Commented] (MESOS-314) Support the cgroups 'cpusets' subsystem.

2016-02-19 Thread Timothy Chen (JIRA)

[ 
https://issues.apache.org/jira/browse/MESOS-314?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15154830#comment-15154830
 ] 

Timothy Chen commented on MESOS-314:


Not sure how it's going to be advertised yet, but I imagine a scheduler can 
just pick one if users don't have a preference.

> Support the cgroups 'cpusets' subsystem.
> 
>
> Key: MESOS-314
> URL: https://issues.apache.org/jira/browse/MESOS-314
> Project: Mesos
>  Issue Type: Story
>Reporter: Benjamin Mahler
>  Labels: twitter
>
> We'd like to add support for the cpusets subsystem, in order to support 
> pinning to cpus.
> This has several potential benefits:
> 1. Improved isolation against other tenants, when given exclusive access to 
> cores.
> 2. Improved performance, if pinned to several cores with good locality in the 
> CPU topology.
> 3. An alternative / complement to CFS for applying an upper limit on CPU 
> usage.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)


[jira] [Commented] (MESOS-314) Support the cgroups 'cpusets' subsystem.

2016-02-19 Thread Casey Sybrandy (JIRA)

[ 
https://issues.apache.org/jira/browse/MESOS-314?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15154692#comment-15154692
 ] 

Casey Sybrandy commented on MESOS-314:
--

Regarding the use of cpuset, would it be possible to have the executor 
automatically pick a CPU to use on a per executor basis?  We have a number of 
containers we run in mesos and we'd love to be able to pin each running 
container to a single CPU.  Which CPU doesn't matter and we will be running 
multiple instances of the same container on a single node, so explicitly 
setting a value won't work for us.

> Support the cgroups 'cpusets' subsystem.
> 
>
> Key: MESOS-314
> URL: https://issues.apache.org/jira/browse/MESOS-314
> Project: Mesos
>  Issue Type: Story
>Reporter: Benjamin Mahler
>  Labels: twitter
>
> We'd like to add support for the cpusets subsystem, in order to support 
> pinning to cpus.
> This has several potential benefits:
> 1. Improved isolation against other tenants, when given exclusive access to 
> cores.
> 2. Improved performance, if pinned to several cores with good locality in the 
> CPU topology.
> 3. An alternative / complement to CFS for applying an upper limit on CPU 
> usage.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)


[jira] [Commented] (MESOS-314) Support the cgroups 'cpusets' subsystem.

2016-01-11 Thread Niklas Quarfot Nielsen (JIRA)

[ 
https://issues.apache.org/jira/browse/MESOS-314?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15091904#comment-15091904
 ] 

Niklas Quarfot Nielsen commented on MESOS-314:
--

Also, I have a vague memory of us having an existing ticket for cpu set 
isolators. [~jvanremoortere] do you recall which ticket it is? I haven't been 
able to dig it up. In terms of the oversubscription work going on in Serenity, 
it would be great to get some notion of core pinning / affinity in Mesos :)

> Support the cgroups 'cpusets' subsystem.
> 
>
> Key: MESOS-314
> URL: https://issues.apache.org/jira/browse/MESOS-314
> Project: Mesos
>  Issue Type: Story
>Reporter: Benjamin Mahler
>  Labels: twitter
>
> We'd like to add support for the cpusets subsystem, in order to support 
> pinning to cpus.
> This has several potential benefits:
> 1. Improved isolation against other tenants, when given exclusive access to 
> cores.
> 2. Improved performance, if pinned to several cores with good locality in the 
> CPU topology.
> 3. An alternative / complement to CFS for applying an upper limit on CPU 
> usage.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)


[jira] [Commented] (MESOS-314) Support the cgroups 'cpusets' subsystem.

2016-01-11 Thread Niklas Quarfot Nielsen (JIRA)

[ 
https://issues.apache.org/jira/browse/MESOS-314?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15091873#comment-15091873
 ] 

Niklas Quarfot Nielsen commented on MESOS-314:
--

[~tstclair] Reg. the cpuset bug in groups. Do you know if this is still the 
case?

> Support the cgroups 'cpusets' subsystem.
> 
>
> Key: MESOS-314
> URL: https://issues.apache.org/jira/browse/MESOS-314
> Project: Mesos
>  Issue Type: Story
>Reporter: Benjamin Mahler
>  Labels: twitter
>
> We'd like to add support for the cpusets subsystem, in order to support 
> pinning to cpus.
> This has several potential benefits:
> 1. Improved isolation against other tenants, when given exclusive access to 
> cores.
> 2. Improved performance, if pinned to several cores with good locality in the 
> CPU topology.
> 3. An alternative / complement to CFS for applying an upper limit on CPU 
> usage.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)


[jira] [Commented] (MESOS-314) Support the cgroups 'cpusets' subsystem.

2014-08-09 Thread Niklas Quarfot Nielsen (JIRA)

[ 
https://issues.apache.org/jira/browse/MESOS-314?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14091755#comment-14091755
 ] 

Niklas Quarfot Nielsen commented on MESOS-314:
--

Any reason not to use `taskset` for cpu pinning? Exposing the available cpus as 
a set i.e. ("core0", ..., "coreN") and running the command with task set (or 
applying it after the fact) seems straight forward. How to tie affinity to the 
cpushare you get may be a messy thing - haven't thought that through.

> Support the cgroups 'cpusets' subsystem.
> 
>
> Key: MESOS-314
> URL: https://issues.apache.org/jira/browse/MESOS-314
> Project: Mesos
>  Issue Type: Story
>Reporter: Benjamin Mahler
>  Labels: twitter
>
> We'd like to add support for the cpusets subsystem, in order to support 
> pinning to cpus.
> This has several potential benefits:
> 1. Improved isolation against other tenants, when given exclusive access to 
> cores.
> 2. Improved performance, if pinned to several cores with good locality in the 
> CPU topology.
> 3. An alternative / complement to CFS for applying an upper limit on CPU 
> usage.



--
This message was sent by Atlassian JIRA
(v6.2#6252)