On Apr 21, 2006, at 11:08 AM, Jeffrey B. Layton wrote:
> I'm using PBS so here is the basics of what my script does:
>
> NET=tcp
> echo $PBS_NODEFILE > machines.$PBS_JOBID
> NP=`wc -l machines.$PBS_JOBID | awk '{print $1}'`
>
> lamboot -b -v -ssi rpi $NET $PBS_NODEFILE
> lamnodes
> mpirun -v -np $NP ./${EXE} >> OUTPUT
> lamhalt -v
>
> Does this seem reasonable? (If it does I'll post some of the errors
> I'm getting in a follow-up email).
Try this one instead... the -ssi rpi parameter only works for
mpirun, not lamboot.
NET=tcp
echo $PBS_NODEFILE > machines.$PBS_JOBID
NP=`wc -l machines.$PBS_JOBID | awk '{print $1}'`
lamboot -b -v $PBS_NODEFILE
lamnodes
mpirun -v -np $NP -ssi rpi $NET ./${EXE} >> OUTPUT
lamhalt -v
By the way, the following should be nearly equivalent.
NET=tcp
lamboot -b -v $PBS_NODEFILE
lamnodes
mpirun -v C -ssi rpi $NET ./${EXE} >> OUTPUT
lamhalt -v
The C option will run on all available "slots", a nice short-cut for
situations like this. If you have built LAM/MPI with TM support, you
could also drop the $PBS_NODEFILE option to lamboot.
Hope this helps,
Brian
--
Brian Barrett
LAM/MPI developer and all around nice guy
Have a LAM/MPI day: http://www.lam-mpi.org/
|
