Loading...

mauiusers@supercluster.org

[Prev] Thread [Next]  |  [Prev] Date [Next]

Re: [Mauiusers] Maui vs pbs_sched Garrick Staples Mon Nov 05 00:20:46 2007

On Wed, Oct 31, 2007 at 12:15:44PM +0200, Alexander Piavka alleged:
> 
>   Hi Garrick,
> 
> >When using maui, you can either work in "ncpus" or "nodes".  Mixing and
> >matching doesn't work.
> 
>  I can't find info on ncpus and nodect both in pbs man pages and torque 
>  wiki.
> Can you you please elaborate on their meaning and how they should be used?

ncpus is interpreted as the number of CPUs *per node* by maui.

nodect is the count of nodes requested or assigned to the job.  It is parsed by
the "nodes" request.  The fact that "nodes" is a string and "nodect" is an
integer is important.  (strings can't be compared to determine if one is
greater).

 
> I've total 55 nodes with mixed number of cpus/cores - 1 ,2 ,4 and 8 in 
> total 113 cpus. So I put the following settings:
>       max_running = 113
>         resources_max.nodes = 55
>         resources_default.nodes = 1:ppn=1
>         resources_available.nodes = 55

Don't set available because pbs_server already knows what is available (there
is 1 corner case where you want to lie to pbs_server).  Use "nodect" for
min/max.  "max.nodes" is meaningless.


> >For your uses, use resources_min/max.nodect and resources_default.nodes.  
> >Keep
> >in mind that "nodes" is a string and doesn't work in min/max comparisons.
> >"nodect" is an integer for comparisons.
> >
> does that mean i heen to change it to:
>         max_running = 113
>         resources_max.nodect = 55
>         resources_default.nodes = 1:ppn=1
>         resources_available.nodect = 55

Unset the resources_available.  Otherwise, yes.


> or it is better to have something like, to have same effect:
>         # remove max_running = 113
>         resources_max.ncpus = 113
>         resources_default.nodes = 1:ppn=1
>         resources_available.ncpus = 113
> or even:
>         # remove max_running = 113
>         resources_max.ncpus = 113
>         resources_default.ncpus = 1
>         resources_available.ncpus = 113

No, those would be mixing ncpus and nodes.  Don't do that.

_______________________________________________
mauiusers mailing list
[EMAIL PROTECTED]
http://www.supercluster.org/mailman/listinfo/mauiusers