[torqueusers] load balancing

Mahmood Naderan nt_mahmood at yahoo.com
Fri Jun 15 04:02:05 MDT 2012


>This is because its a server parameter, see here: 

>http://www.clusterresources.com/torquedocs21/a.bserverparameters.shtml
Thanks.

>As far as I know, it only works if you use pbs_sched.
It is not stated in the document? Did you try?


// Naderan *Mahmood;


----- Original Message -----
From: André Gemünd <andre.gemuend at scai.fraunhofer.de>
To: Mahmood Naderan <nt_mahmood at yahoo.com>; Torque Users Mailing List <torqueusers at supercluster.org>
Cc: 
Sent: Friday, June 15, 2012 1:26 PM
Subject: Re: [torqueusers] load balancing

This is because its a server parameter, see here: http://www.clusterresources.com/torquedocs21/a.bserverparameters.shtml

As far as I know, it only works if you use pbs_sched. Otherwise you need a similar setting in your scheduler, e.g. Maui. 
For the latter, it should be NODEALLOCATIONPOLICY, but afaik there is no predefined setting that is similar. You'll have
to experiment with CPULOAD or PRIORITY. 

A good example from the Maui docs is this:

Example 1: Favor the fastest nodes with the most available memory which are running the fewest jobs.

NODEALLOCATIONPOLICY PRIORITY
NODECFG[DEFAULT] PRIORITYF='SPEED+ .01 * AMEM - 10 * JOBCOUNT' 

Greetings
André

----- Ursprüngliche Mail -----
> >qmgr -c 'set server node_pack = false'
>  Thanks for your help. I can not find documentation about this
>  parameter in "4.1 queue configuration"
>  http://www.clusterresources.com/torquedocs21/4.1queueconfig.shtml
> 
> 
> 
> // Naderan *Mahmood;
> 
> 
> ----- Original Message -----
> From: Andrew Savchenko <bircoph at gmail.com>
> To: Mahmood Naderan <nt_mahmood at yahoo.com>; Torque Users Mailing List
> <torqueusers at supercluster.org>
> Cc:
> Sent: Thursday, June 14, 2012 11:59 PM
> Subject: Re: [torqueusers] load balancing
> 
> Hello,
> 
> On Thu, 14 Jun 2012 10:53:39 -0700 (PDT) Mahmood Naderan wrote:
> > Hi
> > Assume there are two nodes:
> > w1 np=12
> > w2 np=12
> > 
> > When I submit 8 jobs, all 8 jobs are run on w2. It is more
> > desirable to have two 50% load on each node rather than a 100%
> > load on one. The 8 jobs are all single threaded not MPI. Is there
> > any way to define such policy?
> > http://www.supercluster.org/mailman/listinfo/torqueusers
> 
> qmgr -c 'set server node_pack = false'
> should help you.
> 
> Best regards,
> Andrew Savchenko
> 
> _______________________________________________
> torqueusers mailing list
> torqueusers at supercluster.org
> http://www.supercluster.org/mailman/listinfo/torqueusers
> 

-- 
André Gemünd
Fraunhofer-Institute for Algorithms and Scientific Computing
andre.gemuend at scai.fraunhofer.de
Tel: +49 2241 14-2193
/C=DE/O=Fraunhofer/OU=SCAI/OU=People/CN=Andre Gemuend



More information about the torqueusers mailing list