Novel Cooling User Guide

Hartree Centre Logo

Back to Contents Page

Using the Hartree EEC Novel Cooling Demonstrator "Neale" from ClusterVision

This system comprises 120 compute nodes immersed in oil for energy efficient cooling. Some energy and power data is available (details to follow) and the nodes are currently Intel Ivybridge. Therefore use of the system follows the typical pattern for x86 clusters, using a batch system (in this case SLURM) to submit jobs.

Given this is a Hartree EEC Research Machine, there is less support than on Hartree Production Machines (such as Scafell Pike).

Novel Cooling System

Using SLURM

The NCD system uses the SLURM batch scheduler. For information about SLURM see external link: http://slurm.schedmd.com . MPI jobs should be compiled in such a way that the are closely coupled and can be invoked with the "srun" command. This can be achieved in two ways. There is a useful user guide here external link: https://computing.llnl.gov/tutorials/linux_clusters/#MPI .

Linking to libpmi

Use environment variable

This is now included in the intel-mpi module.

SLURM Commands

All the commands have man pages and accept the "--help" option.

LLNL have nice help pages about SLURM: external link: https://computing.llnl.gov/linux/slurm/

To get an interactive shell on node xxx do "srun -w nodexxx -u bash -i".

Sample SLURM Job Script

This was used to run the IMB benchmark. Submit the script with something like "sbatch -n 32 slurm.sh" to run 32 MPI ranks on 32 cores.

Note, if there are any typos in the script, sbatch may simply not return any output.