HPC for numerical methods and data analysis

MATH-505

Information on using the parallel cluster

This page is part of the content downloaded from Information on using the parallel cluster on Wednesday, 25 December 2024, 18:23. Note that some content and any files larger than 50 MB are not downloaded.

Description

Information on using the parallel cluster

Page content

Cluster details:

Cluster name: helvetios (cpu)

Account:  math-505

Log in (while on EPFL network, directly or through vpn): 

  • Generate for your first connection a pair of public/private key using ssh-keygen (optional)
  • Log in: ssh <username>@helvetios.hpc.epfl.ch

Loading modules for using Python with MPI:

  • Run the command "module load gcc openmpi python py-mpi4py" to be able to run with tools compiled and optimized for the hardware by Scitas
Running a code:
Example of a script that uses srun to execute a code with 38 MPI ranks over two nodes and 1 thread per rank. It asks for 7000 MB of RAM (this is defined per node, so in total the job gets 14'000 MB), 20 minutes for the job and asks for the parallel QOS. The parallel QOS is needed for any job using more than 1 node and at least 37 cores, the default QOS is serial which is up to all 36 cores of a node:
#!/bin/bash -l
#SBATCH --nodes=2
#SBATCH --ntasks-per-node=19
#SBATCH --cpus-per-task=1
#SBATCH --mem=7000
#SBATCH --time=20:00
#SBATCH --qos=parallel
#SBATCH --account=math-505
module load gcc openmpi python py-mpi4py
srun ./my_python_script.py

Another example of a script example.batch
sbatch example.batch

Additional links: