[Reta]

Hi

just wondering if anybody could help me with a small problem i'm having.
I'm running star in batch mode but anytime I go over np=16, which are the number of cores per machine. I oversubscribe the process wasting away the other nodes I have access to since i focuses on the

From what I have researched its not a problem with the the submission file but perhaps with how I have the simulation itself setup.


#!/bin/sh
# Grid Engine options (lines prefixed with #$)
#$ -N HighRes ##Rename job
#$ -cwd ##Directory where the job is run
#$ -pe sharedmem 32
#$ -l h_rt=8:00:00
#$ -l h_vmem=16G
#$ -M s1403217@sms.ed.ac.uk ## Email when the job begins and starts
#$ -m beas
#$ -j y
# These options are:
# job name: -N
# use the current working directory: -cwd
# runtime limit of 5 minutes: -l h_rt
# memory limit of 1 Gbyte: -l h_vmem

# Initialise the environment modules
. /etc/profile.d/modules.sh

# Load Star
module load eng/starccm+/11.06
starccm+ -batch SimpleRun.java -np 32 HighResV2@meshed.sim


thats the kind of submission script i have with just a difference of using mpi instead of sharedmem and loading the mpi with a module load.

Thanks for any help, I feel like its just a simple solution im missing here.

[me3840]

You may have to specify the nodelist with the -machinefile flag.