[torquedev] nodes, procs, tpn and ncpus

[Michel Béland]

Mgr. Šimon Tóth wrote:

>>> nodes=x will mean the same as procs=x.
>> I don't think this should be the case... Moab reinterprets it to mean
>> the same thing, but historically with PBS that is not how has been
>> interpreted.
> 
> Plus PBSPro has a strong distinction between processes and cpus.
> 
> -l nodes=2:ppn=3:ncpus=4
> 
> - you get two nodes with 3 processes on each node, and each process will
> get 4 cpus

If I remember how PBS Pro worked when we used to use it, the distinction 
was only to set $PBS_NODEFILE and $OMP_NUM_THREADS (and $NCPUS). In turn 
it did have an effect on the processes and threads used. When the 
introduced the -lselect syntax, I believe that they screwed it up a 
little bit, with the introduction of mpiprocs and ompthreads. I am not 
sure that I remember precisely what was wrong and have no way to test 
it, but I think that the problem was that you had to specify mpiprocs, 
ompthreads and ncpus insinde the -lselect even if ncpus could be deduced 
from mpiprocs and ompthreads.

Anyway, I do not know if we want to import the -lselect syntax as is 
from PBS Pro, but maybe it would be a nice idea to get rid of its 
imperfections.

People are saying on the mailing list that -lselect has to be 
implemented, but in PBS Pro, its flexibility comes from its twin 
resource called -lplace. This is where you tell PBS how to allocate the 
virtual processors on the (virtual) nodes.

-- 
Michel Béland, analyste en calcul scientifique
michel.beland at rqchp.qc.ca
bureau S-250, pavillon Roger-Gaudry (principal), Université de Montréal
téléphone : 514 343-6111 poste 3892     télécopieur : 514 343-2155
RQCHP (Réseau québécois de calcul de haute performance)  www.rqchp.ca