.

Arcus Phase B

Arcus Phase B information

Here are details for access and use of the new system.

ARC users can connect to the new cluster via ssh to:

arcus-b.arc.ox.ac.uk

using standard ARC username credentials.  Please email support@arc.ox.ac.uk if you have any problems connecting to the system.

Arcus Phase B is a HPC cluster consisting of Dual Haswell CPU nodes which have 16 cores per node and a minimum of 64Gb of memory. Software is available via modules. If you don't find software you require in modules, then please ask for it.

Job submission is managed by the SLURM job manager not Torque. You will have to change your submission scripts. Fortunately, there is almost a one-to-one mapping from Torque to SLURM. Thus, this script that works on Arcus

#!/bin/bash

#PBS -V
#PBS -l nodes=2:ppn=16
#PBS -l walltime=100:00:00
#PBS -N QEtest

cd $PBS_O_WORKDIR/
. enable_arcus_mpi.sh mpirun $MPI_HOSTS pw.x -in ausurf.in &> ausurf.out

maps to this script for Arcus-b:

#!/bin/bash

#SBATCH --nodes=2
#SBATCH --ntasks-per-node=16 #SBATCH --time=100:00:00
#SBATCH --job-name=QEtest . enable_arcus-b_mpi.sh mpirun $MPI_HOSTS pw.x -in ausurf.in &> ausurf.out

There are two differences: a) no need to propagate the environment (#PBS -V) and b) no need to go to the working directory. SLURM does both automatically.

Say the above script is called ausurf.sh. You submit the job with

sbatch ausurf.sh

You monitor jobs with squeue or better, squeue -u <username> and kill a job with scancel 123. Output from SLURM is in file slurm-123.out after the job finishes. (Of course, 123 is to be replaced with the ID of the job.)