Work on the LIP6 GPU Cluster: A Quick Guide

Here’s a compact guide to help you work on the LIP6 GPU cluster: from connecting, to running jobs, to setting up interactive Jupyter Notebooks.

• LIP6 Convergence Official Instructions

Connecting to LIP6 Clusters

To streamline your SSH connection, add these settings to your ~/.ssh/config:

Host barder.lip6.fr
    User <username>
    ForwardAgent yes

Host cluster.lip6.fr
    User <username>
    ProxyJump barder.lip6.fr

Then simply connect:

• For CPU cluster:

  ssh cluster.lip6.fr
  

• For GPU cluster:

  ssh front.convergence.lip6.fr
  

No need to remember complex jump commands anymore!

Running Jobs on the GPU Cluster

Once connected, you’ll interact with the SLURM job scheduler. Here’s the basic flow:

• Check the queue

  squeue
  

• Submit a job
  Prepare a run.sh script and submit it:

  sbatch run.sh
  

• Cancel a job

  scancel <job_id>
  

Example Job Scripts

1. Running a Python Script (script.sh)

Here’s a simple example of a job that runs a Python script:

#!/bin/bash
#SBATCH --job-name=test_script
#SBATCH --nodes=1
#SBATCH --gpus-per-node=1
#SBATCH --time=1440
#SBATCH --mail-type=ALL
#SBATCH --output=%x-%j.out
#SBATCH --error=%x-%j.err

module purge
module load python/anaconda3
eval "$(conda shell.bash hook)"
conda activate myenv
python main.py --test_nbr_workers=0

Just modify main.py with your actual Python script name.

2. Running a Jupyter Notebook (notebook.sh)

Want to start a Jupyter notebook? Here’s the script:

#!/bin/bash
#SBATCH --job-name=test_jupyter
#SBATCH --nodes=1
#SBATCH --gpus-per-node=1
#SBATCH --time=60
#SBATCH --mail-type=ALL
#SBATCH --output=%x-%j.out
#SBATCH --error=%x-%j.err

module purge
module load python/anaconda3
eval "$(conda shell.bash hook)"
conda activate myenv
jupyter notebook

Then on your local machine, use SSH port forwarding:

ssh -J front.convergence.lip6.fr -L 8888:localhost:8888 node01.convergence.lip6.fr

Or if you’re inside the cluster front-end or VSCode terminal:

ssh -L 8888:localhost:8888 node01.convergence.lip6.fr

Now you can open http://localhost:8888 on your browser to access your notebook remotely!

Interactive Sessions

Need an interactive terminal on a GPU node?

srun --nodes=1 --ntasks-per-node=1 --time=01:00:00 --pty bash -i

This is great for quick testing or troubleshooting before submitting longer jobs.