[Mauiusers] qtop: yet another tool to struggle with torque and PBS family systems (and soon more)

Fotis Georgatos fotis at cern.ch
Wed Jul 6 03:36:23 MDT 2011


If anybody of you is applying maui over an SGE instance,
you may be interested in this tool: http://cern.ch/fotis/QTOP
especially, if you struggle with job scheduling issues across nodes.
It seems that confining maui policies is an upcoming sport. ;-)

I recently hacked together an sge2pbs wrapper for sites using SGE;
that wrapper is still experimental but seems to be working fine so far
at a certain site. Contact back in private for getting a copy and test it.

It's particularly interesting to receive feedback from people
who have experience with applying maui to both PBS ans SGE systems.

fyi. with PBS, you are only 1 rpm command away from using qtop


On 16/10/2010 11:52, Fotis Georgatos wrote:
> Hi,
> since maui had been THE cause for starting writing qtop shell script
> some years ago, I forward the below appended message here, too.
> btw.
> If someone of you is aware on any kind of hard guarantees of scheduling
> with maui or else, it would be very interesting to hear your feedback.
> What I am specifically looking for, are computer science clauses of the type:
> "given such and such algorithm in the scheduler, max queue delay is so much"
> ie. the scheduling building block for Urgent Computing applications.
> If, any of you has hands-on experience on the matter, please contact back.
> thank you in advance for any answer,
> Fotis
> -------- Original Message --------
> Subject: Fwd: qtop: yet another tool to struggle with torque and PBS family
> systems
> Date: Fri, 15 Oct 2010 20:56:42 +0300
> From: Fotis Georgatos <fotis at cern.ch>
> To: Torque Users Mailing List <torqueusers at supercluster.org>
> Hi,
> thank you to all the people who provided feedback since 1st release of qtop;
> now a new release is out with an rpm package available,
> and multiple pending requests have been addressed:
> * fixed visual output for sites with 1000s of cores/jobs
> * highlighting of queues in the header (coloring of queue names, like jobs)
> * added command-line options for most of the tunable parameters
> * separately report Running + Queued jobs.
> * provide a default /etc/qtop.conf
> * an example is provided on how to convert the output of qtop in a colored
> web-page
> (and if you see funny output rendering in your smartphone, well that's a
> WebKit bug ;-)
> http://cern.ch/fotis/QTOP
> cheers,
> Fotis
> ps.
> I'm using the list communication channel out of convenience, and thanks for that.
> -------- Original Message --------
> Subject: qtop: yet another tool to struggle with torque and PBS family systems
> Date: Wed, 01 Sep 2010 01:20:29 +0200
> From: Fotis Georgatos <fotis at cern.ch>
> Organization: CERN
> To: torqueusers at supercluster.org
> CC: Fotis Georgatos <fotis at cern.ch>
> Hi,
> it is not uncommon for shepherds of PBS-family based clusters to wander around
> in the system, trying to understand where users' jobs and site resources graze.
> Or, you may just try to understand if you are being hit by something like a
> bug. (*)
> Fortunately,
> you are not alone in this world and others have same troubles as you do ;-).
> Here is a script I wrote to try to get better control over any torque or pbs
> instance:  http://cern.ch/fotis/QTOP
> It provides a brief summary of your PBS and Nodes status, along with a job
> matrix.
> In case you ask, the CPUids are the ones reported at command pbsnodes -a,
> so you may wish to check that its output seems reasonable, before trying qtop.
> It can be particularly useful if you assign colors to your scheduler's policy
> groups,
> so that you can visually check if your policy is honored; be prepared for
> surprises.
> Generally, I hope it can help you to keep the entropy of a torque system low,
> by giving a fast overview of what is going on.
> enjoy,
> Fotis
> (*)
> http://www.supercluster.org/pipermail/torqueusers/2010-August/011198.html

echo "sysadmin know better bash than english" | sed s/min/mins/ \
	| sed 's/better bash/bash better/' # Yelling in a CERN forum

More information about the mauiusers mailing list