OK, some more looking may have turned up what we want.
PBS supports the concept of "dedicated time". These are specified in
/usr/pbs/etc/pbs_dedicated. During these times, only specified queues can
run jobs and I don't think we have any queues set this way. So, from the
documentation, if we set a dedicated time, and then re-start the scheduler,
no jobs would get scheduled during this time. Looking at the currently
running jobs, it looks like all of the currently running jobs should be
complete in 225 hours - 9 days and 9 hours - about first thing in the
morning on Sat. Oct. 7.
Dave
-----Original Message-----
From: Research Computing Support [mailto:[log in to unmask]] On
Behalf Of David Woods
Sent: Wednesday, September 27, 2006 2:15 PM
To: [log in to unmask]
Subject: Re: mulnx33 unresponsive
I was thinking of a way to specify when the nodes would be down, but still
allow jobs to run as long as the requested time can be completed before the
scheduled downtime. I Googled this some more and this may be a feature of
the Maui scheduler, and not part of PBS.
Dave
-----Original Message-----
From: Research Computing Support [mailto:[log in to unmask]] On
Behalf Of jaime combariza
Sent: Wednesday, September 27, 2006 10:07 AM
To: [log in to unmask]
Subject: Re: mulnx33 unresponsive
one way is to stop all queues so no jobs will start until the system
is ready and queues are started again.
You can still submit jobs but these will not run (if there are slots
available)
I think that you can also disable all queues if you do not want
anyone to submit jobs
At 08:34 AM 9/27/2006, you wrote:
>Does PBS have a way to schedule downtime so that queued jobs that can't
>complete before the downtime won't start? I thought it did, but can't find
>it in the documentation.
>
>Dave
>
>-----Original Message-----
>From: Research Computing Support [mailto:[log in to unmask]] On
>Behalf Of jaime combariza
>Sent: Wednesday, September 27, 2006 8:06 AM
>To: [log in to unmask]
>Subject: Re: mulnx33 unresponsive
>
>That is fine. Let me know the best day (as soon as posisble) to give
>the users enough time to prepare.
>
>As far as the HPL test under GigE, I'd like to predict efficiency
>between 0.5 -0.6, so for our system
>we can get 0.89 TFLOps (assuming 0.58 eff.) Infiniband gives us close
>to 0.84 eff.
>
>
>At 05:28 PM 9/26/2006, you wrote:
> >All,
> >
> >We have mulnx33 (one of our segment servers) unresponsive today. I
> >power cycled it.
> >Hopefully, no jobs are affected.
> >
> >I think we'll request one full day for the patch. Just so that I can
> >run hardware diag as well. I think it has happened several times to
>mulnx33.
> >
> >Thanks,
> >------------
> >Robin
> ><mailto:[log in to unmask]>[log in to unmask]
> >(513) 529-1483
> >
> >"Academia politics is the most vicious precisely because the stake
> >is so small" - Kissinger
> >
> >
>
>
>_______
>Jaime E. Combariza, Ph.D.
>Assistant Director Research Computing
>http://www.muohio.edu/researchcomputing
>Miami University
>(513) 529-5080
_______
Jaime E. Combariza, Ph.D.
Assistant Director Research Computing
http://www.muohio.edu/researchcomputing
Miami University
(513) 529-5080
|
