[Moabusers] Nice - priority in operating system

[Lukasz Skital]

Hi,

This is mostly maui question, because it is widely used in our
project. However, we are also considering moab for  our site.

Thank you for the answer. Does anybody know if redefining 'suspend' to
'low priority via nice' is possible with maui?

Best regards,
Lukasz

On 3/2/07, Dave Jackson <jacksond at clusterresources.com> wrote:
> Lukasz,
>
>   Is this a Maui question or a Moab question?  With Moab this can be
> done by redefining the 'suspend' state to mean 'low priority via nice'.
> With Maui, I don't know if the needed infrastructure is there.
> Regarding the TORQUE component, I believe the infrastructure is there
> but would need to verify it is dynamic.
>
> Dave
>
> On Fri, 2007-03-02 at 09:34 +0100, Lukasz Skital wrote:
> > Hi *,
> >
> > I am new to the list so I introduce my self a little: I am
> > cluster/grid administartor for EGEE project (http://www.eu-egee.org/).
> > I also works in few other grid related projects.
> >
> > One of the project I am involved,  could benefit from following feature:
> >
> > Preempting mechanism, which, instead of
> > suspending/killing/requeuing/checkpointig is altering process priority
> > in operating system (in linux you can use renice to do this).
> >
> > Is it possible to be done with toruqe/maui? Is it possible with other
> > resource management system?
> >
> > If you are not sure, what exactly I request, there is a little background:
> >
> > We have general purpose grid infrastructure (gLite based) with torque
> > and maui on clusters. We have some specific soft real-time application
> > which need to access resources as quick as it is possible, but we want
> > to avoid static reservations. So the idea is to preempt working low
> > priority job by our high priority jobs. However, we do not want to
> > kill or suspend low priority jobs (checkpointing is not supported),
> > but instead want to renice it, to serve most of CPU cycles high
> > priority job and the low priority job will maintain its operations
> > (esp. keeps network connections). So instead of  suspending, the job
> > is put in to low priority (renice 20) for the time, when high prio job
> > is running on the same CPU.
> >
> > Memory constraints is not the case, because our high priority jobs
> > have low memory requirements.
> >
> > Best regards,
> > _______________________________________________
> > moabusers mailing list
> > moabusers at supercluster.org
> > http://www.supercluster.org/mailman/listinfo/moabusers
>
>


-- 
Łukasz Skitał   <l.skital at cyfronet.pl>  GG: 1279114