How do I run a parallel job on our cluster (ulgbcm)?

Below is a simple script to run a parallel job on 4 nodes on our cluster. Use

qsub play.pbs

to submit the script.

An example script

#PBS -N solver
#PBS -l walltime=0:30
#PBS -o /users/mcneile/playground/mpi/log_mpi
#PBS -e /users/mcneile/playground/mpi/log_mpi_e
#PBS -l nodes=4:score
#
#  Simple script to run a parallel enviroment
#  on our cluster
#
# The executable
# ${HOME}/playground/mpi/a.out
# is run in the ${HOME}/playground/mpi directory.

cd ${HOME}/playground/mpi

echo "Start of the job at `date`"


echo "PBS_NODEFILE ${PBS_NODEFILE}"

# The variable PBS_NODEFILE seems to be set
# by the environment.

scout -wait -F ${PBS_NODEFILE} -e scrun -nodes=4 ${HOME}/playground/mpi/a.out

echo "End of the job at `date`"
exit

Additional documentation

The main source of documentation is online at

netscape   /opt/score/doc/index.html.en

(on ulgbcm). The The Streamline Computing Cluster Guide contains useful information. The parallel system is built by the PC Cluster Consortium.

FAQ home page

Running the MILC code

The MILC code is the most portable lattice QCD codes ever written. As proof of this, to compile on ulgbcm, you need to modify: Make_linux_mpi

CC               = mpicc

Then compile using the instructions supplied by MILC.

The MILC code reads the input files from standard input. The way standard input is read on ulgbcm is brain dead to say the least. The following seems to work.

scout -wait -F ${PBS_NODEFILE} -e scrun -nodes=4 scatter == ${HOME}/milc/clover_hybrids/${exec} < ${HOME}/milc/clover_hybrids/${input}

Last modified: Mon Nov 17 15:02:50 2003