[Mauiusers] Fairness Policy Configuration

Greenseid, Joseph M (IS) Joseph.Greenseid at ngc.com
Fri Feb 26 07:38:48 MST 2010


Marcelo,
 
Have you tried setting a CLASSCFG directive for the queue you're trying to manage?  For example, we have a group that is only allowed to have a max number of idle nodes of 64 in the queue "work", and that looks like this:
 
CLASSCFG[work] MAXINODE[GROUP:group1]=64
 
--Joe

________________________________

From: mauiusers-bounces at supercluster.org on behalf of Marcelo Zimmer
Sent: Thu 2/25/2010 4:18 PM
To: mauiusers at supercluster.org
Subject: Re: [Mauiusers] Fairness Policy Configuration


Hi,

Thank you for your prompt response, but the suggestions don't seem to be solving my problem. Actually, I've tracked the submitted jobs with the "showq" command and I couldn't see any blocked ones even if the number of idle jobs was greater than the number I wast trying to specify (actually, all non-running jobs appeared with the flag "Q"). As I said in the previous email, my aim is to set a quite soft policy, which is basically allowing users to use as many available processors as possible (in an extreme case, a single user can even take control of the whole cluster and use all processors at once), while precluding users to submit an extesive number of jobs that stay on "Q" mode, which would result in people "taking control" of the cluster during on month, for instance. As result, I found 5 a good number to set as the maximum number of idle jobs per user.

This way, as far as I could understand, the number specified in "max_user_queuable" or "max_queuable" flags in torque (PBS) indicates the maximum number of jobs a user can put in queue, regardless of its status (either running or idle). For that, I found very hard to set a number for those flags, because they would result either in unused processors while there's just one person in need to use the cluster or in setting a high value that would allow user to keep a long list of idle jobs and preventing (or delaying someone else's job to start) others to use the cluster.

Do I still missing anything?

Best regards,

Marcelo Zimmer

-- 
===================================================
                               Marcelo Zimmer S. Flores
                                 (PhD Physics Student)

Address:
       Universidade Estadual de Campinas - UNICAMP
       Instituto de Fisica Gleb Wataghin - IFGW
       Grupo de Solidos Organicos e Novos Materiais - GSONM
       Departamento de Fisica Aplicada (DFA) - Room 61
       Campinas - Sao Paulo - Brazil
       13083-970
       Caixa Postal: 6165
       zimmer at ifi.unicamp.br
===================================================

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.supercluster.org/pipermail/mauiusers/attachments/20100226/8a23026c/attachment.html 


More information about the mauiusers mailing list