[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Condor-users] MPI and multiple SMP machines



> Jens Harting wrote:
>> No, unfortunately not. And this is independent of the entry being in the
>> hostname.local file or in condor_config, as well as putting the name of the
>> group in quotes or not. But in princuiple it should show up, shouldn't it?
>
> Ah, is ParallelSchedulingGroup in the StartdExprs list on the startd? It needs
> to be there, but it looks like the manual doesn't mention this. 
> I'll fix the manual.

That was it!

I have added this line to the local config file:

STARTD_EXPRS = $(STARTD_EXPRS), DedicatedScheduler, ParallelSchedulingGroup

and this one:

ParallelSchedulingGroup = "$(HOSTNAME)"


Now, every SMP machine has its own ParallelSchedulingGroup and it works as 
I hoped. Thank you very much again for your help, Greg.

One last question: can I add more than one ParallelSchedulingGroup on a 
single host? For example, it would be great to have for example one group 
that is given by the hostname and limits 2-4 CPU jobs to only run on a 
single machine. In addition, there are a number of nodes which are pretty 
well connected and could also run MPI jobs across all of them if the 
communication requirements are not too high. It would be great to have a 
second group so that the user can decide if he wants to use that 
opportunity to run jobs across multiple machines or not.

I tried something like

ParallelSchedulingGroup = "$(HOSTNAME),2ndgroup"

but that did not work.

Jens

--------------------------------------------------------------------------
Jens Harting <jens@xxxxxxxxxx>
Werastr. 125, D-70190 Stuttgart, Tel.:+49-711-3589705
Office: ICP,Pfaffenwaldring 27,D-70569 Stuttgart, Tel.:+49-711-685-67609