MARWAN docs NOC knowledge base
DocsHPC

Use the HPC-MARWAN cluster

How researchers in Morocco get an account on the national HPC-MARWAN cluster, log in, store data, load software modules, run Slurm jobs and install their own applications.

Published Saadia Bahti
About this guide

Based on the HPC-MARWAN application install guide (December 2021) and the FAQ of the HPC-MARWAN wiki, brought together and translated to English in November 2026.

What HPC-MARWAN is#

HPC-MARWAN is the national high performance computing cluster run by CNRST for Moroccan universities and research centres. Researchers use it for computations too big or too long for a laptop: simulations, genomics, climate models, machine learning.

The cluster is a group of Linux servers that work together:

Part What you do there
Login node hpc-login.marwan.ma Connect, edit files, prepare and submit jobs. Not for computing.
Compute nodes Run your jobs, assigned by the scheduler.
Slurm The scheduler. It queues jobs and gives each one free resources.
Shared storage Your files, seen the same way from every node.

Get an account#

Access is open to members of the Moroccan academic community who are faculty researchers (enseignants-chercheurs) or PhD students.

  1. Fill in the access request form (PDF, you can type in the fields) and sign it.
  2. Sign the HPC-MARWAN usage charter and scan it.
  3. Email both to hpc@marwan.ma from your institutional address, with:
    • PhD students: your enrolment certificate (attestation d'inscription) for the doctoral programme, for the current academic year;
    • faculty: your work certificate (attestation de travail).

Requests from personal email addresses aren't accepted. The team checks the documents and sends you your login details. An incomplete request is sent back to be completed.

Log in#

You need an SSH client. Linux and macOS have one built in. On Windows use the built-in ssh command in PowerShell, or MobaXterm, which also shows graphical programs.

$ ssh login@hpc-login.marwan.ma

Replace login with your user name. Linux doesn't show anything while you type your password, no dots or stars: type it and press Enter. You are also asked for a verification code from Google Authenticator on your phone.

Can't connect?

Most access problems come from a mistyped password (watch out for extra spaces) or from too many failed attempts, which blocks your public IP address for a while. If the verification code is refused, check that your phone sets its date and time automatically: the codes depend on the exact time. Write to hpc@marwan.ma with a screenshot of the error and your public IP address.

To copy files between your computer and the cluster, use scp (or the file panel of MobaXterm):

$ scp results.tar.gz login@hpc-login.marwan.ma:/data/login/
$ scp login@hpc-login.marwan.ma:/data/login/output.csv .

Where to keep your files#

Directory Size Use it for
/home/login 100 GB Scripts, configuration, small files
/data/login 500 GB Input data, results, your own software
/scratch/users/login 500 GB Temporary files of running jobs

Copy results you need to keep back to your own storage. The cluster is for computing, not for archives.

Run jobs with Slurm#

Work on the cluster in batch mode: prepare your input files, describe the job in a script, and submit it. Slurm puts it in the queue and starts it when resources are free. You can log out meanwhile; check later with squeue and read the output files.

A minimal job script, run.sl:

run.sl
#!/bin/bash
#SBATCH --job-name=my-test
#SBATCH --partition=defq
#SBATCH --nodes=1
#SBATCH --ntasks=8
#SBATCH --time=02:00:00
#SBATCH --output=my-test-%j.out

module load <the modules your program needs>
mpirun ./my_program input.dat

The commands you'll use most:

Command Does
sbatch run.sl Submit the job script
squeue -u $USER Show your jobs, pending and running
scancel 12345 Cancel job 12345
sinfo Show partitions and free nodes
sacct -j 12345 Show what a finished job used

For short tests or post-processing, ask for an interactive shell on a compute node:

$ salloc -N 1 -n 1 -p defq bash

Interactive sessions have to wait for free resources like any job, and end when their time is up. Keep the heavy work for batch jobs.

Partitions. defq is the test partition, limited to 2 hours. For real work pick shortq, mediumq or longq by how long your job runs, and set --time to match. sinfo shows their limits.

Limits per user: 10 running jobs, 10 more waiting in the queue, and 64 CPU cores in total.

Don't compute on the login node

The login node is shared by everyone who is connected. Long or heavy programs started there slow everybody down and may be killed. To stop one you started by mistake:

$ ps -u $USER -o pid,%cpu,cmd
$ kill PID
Scripts written on Windows

A script saved with Windows line endings fails with Batch script contains DOS line breaks. Convert it with dos2unix run.sl, or set your editor (Notepad++, VS Code) to Unix (LF) line endings.

Use the installed software#

Software on the cluster is provided as environment modules. Loading a module sets up the paths for that program and version.

$ module avail              # what can be loaded now
$ module spider gromacs     # search all versions of a program
$ module load gromacs       # load it
$ module list               # what is loaded
$ module purge              # unload everything

Put the same module load lines in your job scripts. If a program you need isn't there, ask hpc@marwan.ma to install it for everyone, or install it yourself as below.

Install your own software#

You can install programs in your own space without administrator rights. Keep them under /data/login/local:

$ mkdir -p /data/$USER/local/{apps,modfiles,share}
Directory Holds
local/share Downloaded source archives
local/apps Installed programs, one directory per program and version
local/modfiles Your own module files, if you write some

Build on a compute node, not on the login node:

$ salloc -N 1 -n 1 -p defq bash

Get the source with wget or git clone from the cluster, or copy it from your computer with scp into /data/login/local/share/. Then build with an install prefix in your space.

For a program that uses configure:

$ cd /data/$USER/local/share
$ tar -xzf myprog-1.2.tar.gz && cd myprog-1.2
$ ./configure --prefix=/data/$USER/local/apps/myprog/1.2
$ make -j 4
$ make install

For a program that uses CMake:

$ mkdir build && cd build
$ cmake .. -DCMAKE_INSTALL_PREFIX=/data/$USER/local/apps/myprog/1.2
$ make -j 4 && make install

Load the compiler and library modules the program needs before you build, and the same ones in your job scripts. To run it, add its bin directory to your path in the job script:

export PATH=/data/$USER/local/apps/myprog/1.2/bin:$PATH

Get help#

Write to hpc@marwan.ma. Say what you ran, the job ID, and the error message. The team helps each researcher individually, including with porting and installing applications.

Further reading#