[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Running PDP++ using multiple processors
Assuming you have mpich installed, here are my notes for running an
interactive session:
===============================================
using mpirun.ch_gm to run interactively (X11)
===============================================
mpirun.ch_gm -np $1 -machinefile machines /usr/local/bin/dm_leabra++ $2
this is designed to be in a shell script w two/args, one that is the number of
processors ($1) and the other that is a project file ($2). the file
'machines' in the current directory specifies the machines to run on. Note
that mpirun.ch_gm is for the myrinet GM version -- you might want to use p4
or smp or whatever you run on your machine..
Also need to make an ~/.ssh/rc file that runs the xauth process only
on non-processing nodes -- otherwise they all complain about
locking the .Xauthority file.
shrthost=${HOSTNAME:0:4}
if [[ "$shrthost" != "node" || "$HOSTNAME" == "node1" ]];
then if read proto cookie;
then echo "add $DISPLAY $proto $cookie" | xauth -iq -;
fi
fi
----------------------
In any case, the key command is some version of mpirun -- check the man pages
for your local install to see how your version works. The batch submission
scripts use mpiexec instead of mpirun, which works better with our pbs batch
scheduler (mpirun is independent of any batch scheduling you might have).
Hope this helps,
- Randy
On Tuesday 13 July 2004 04:34 pm, 02armsts@utsc.utoronto.ca wrote:
> Hello again,
>
> Thank you for your previous advice regarding running sims in batch mode.
> I am now attempting to run the distributed memory binaries on our grid,
> but I am not sure how to specify some settings.
>
> Before figuring out how to submit jobs to remote nodes using qmon, I would
> just like to run the dm_bp++ and dm_leabra++ binaries on the master node
> using multiple processors. Our master node has 4 CPUs. Right now when
> run dm_bp++, "DMEM Running on 1 processors" is displayed. How do I go
> about changing changing this so it will use all 4 processors?
>
> I have reviewed README.dmem, and dm_qsub, but I am still not sure how to
> proceed. In README.dmem, it states that dmem works in either gui or
> non-gui mode, so I am assuming that it is possible to configure the bp and
> leabra binaries to run on the master node using multiple processors, but I
> am not sure how to set this argument. The examples of using the "special"
> scripts all use the -nogui flag and seem designed for submission to a
> queue. Again, any suggestions you have would be greatly appreciated. I
> would be happy to provide an outline of how we set up our system once we
> are finished for other pdp++ users.
>
> Thanks,
>
> Blair Armstrong
> 02armsts@utsc.utoronto.ca
> University of Toronto at Scarborough