Page tree

Versions Compared

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.

...

  1. Load the compiler you want to use and python using modules

    Code Block
    languagebash
    $ module load gcc
    $ module load python


    Code Block
    languagebash
    $ module load intel
    $ module load python


  2. Create a virtual environment

    In the following, we use GCC but the same procedure applies to Intel compiler. In your home folder, create a virtual environment:

    Code Block
    languagebash
    $ virtualenv -p python3 --system-site-packages opt/venv-gcc
    
    Running virtualenv with interpreter /ssoft/spack/arvine/v1/opt/spack/linux-rhel7-skylake_avx512/gcc-8.4.0/python-3.7.7-drpdlwdbo3lmtkcbckq227ypnzno4ek3/bin/python3
    Already using interpreter /ssoft/spack/arvine/v1/opt/spack/linux-rhel7-skylake_avx512/gcc-8.4.0/python-3.7.7-drpdlwdbo3lmtkcbckq227ypnzno4ek3/bin/python3
    Using base prefix '/ssoft/spack/arvine/v1/opt/spack/linux-rhel7-skylake_avx512/gcc-8.4.0/python-3.7.7-drpdlwdbo3lmtkcbckq227ypnzno4ek3'
    New python executable in /home/user/opt/venv-gcc/bin/python3
    Also creating executable in /home/user/opt/venv-gcc/bin/python
    Installing setuptools, pip, wheel...
    done.
    


  3. Activate virtual environment

    Code Block
    languagebash
    $ source opt/venv-gcc/bin/activate
    
    (venv-gcc) [user@izar ~]$


  4. Install Jupyter

    Code Block
    languagebash
    (venv-gcc) [user@izar ~]$ pip install jupyter
    Collecting jupyter
    ..
    ..


  5. Set passwordless access to Izar by using ssh key and have the following in your ~/.ssh/config file on your personal computer (How to generate an SSH key). It is assumed that you are inside the EPFL network (or connected through VPN).

    Code Block
    languagebash
    Host izar
       Hostname izar.epfl.ch
       User [username]
    
    Host i*.izar
       ProxyCommand ssh -W $(echo "%h" | sed -e 's/.izar//'):%p izar
    


  6. Run jupyter notebook on Izar.
    The script below is a template that allows you to start jupyter notebook on a compute node of Izar. You can copy it in a file called launch_jupyter.sh for example. It has to be placed in your home (or you have to modify it accordingly). Note that all the modules that jupyter may need have to be loaded here. As an example, here, we loaded the modules that allow us to use TensorFlow in the notebook.

    Code Block
    languagebash
    titlelaunch_jupyter.sh
    #!/bin/bash -l
    #SBATCH --job-name=ipython-trial
    #SBATCH --nodes=1
    #SBATCH --gres=gpu:1
    #SBATCH --exclusive
    #SBATCH --time=01:00:00
    #SBATCH --output jupyter-log-%J.out
    
    module load gcc mvapich2 py-tensorflow
    
    source opt/venv-gcc/bin/activate
    
    ipnport=$(shuf -i8000-9999 -n1)
    
    jupyter-notebook --no-browser --port=${ipnport} --ip=$(hostname -i)

    Launch your job as usual:

    Code Block
    languagebash
    sbatch launch_jupyter.sh

    Once the job is running analyze the output jupyter-log-[SLURM_ID].out. Then, look for a line like the following:

    Code Block
    languagebash
        Or copy and paste one of these URLs:
            http://10.91.27.63:8504/?token=4b17ae5cfa505b5470dc84bb5240ab43ae714aa9480a163c
    

    It has the form:

    Code Block
    languagebash
    http://<IP ADDRESS>:<PORT NUMBER>/?token=<TOKEN>


  7. On your local machine do the following with the information provided by the above step

    Code Block
    languagebash
    ssh -L <PORT NUMBER>:<IP ADDRESS>:<PORT NUMBER> -l <USERNAME> izar.epfl.ch -f -N
    
    ## Or if you created the ssh config from step 5
    ssh -L <PORT NUMBER>:<IP ADDRESS>:<PORT NUMBER> izar -f -N

    For our example, this gives:

    Code Block
    languagebash
    ssh -L 8504:10.91.27.63:8504 -l user izar.epfl.ch -f -N
    
    ## with config
    ssh -L 8504:10.91.27.63:8504 izar -f -N


  8. Now you should be able to access to Izar compute node through the web browser by pasting the following address

    Code Block
    languagebash
    http://localhost:<PORT NUMBER>/?token=<TOKEN>

    For our example, this gives:

    Code Block
    languagebash
    http://localhost:8504/?token=4b17ae5cfa505b5470dc84bb5240ab43ae714aa9480a163c



Here is a screenshot of a jupyter session running a simple TensorFlow example. The computations are run on the cluster node, but you access the results directly from your browser.

Image Modified

If you wish to use tensorboard you must open a second connection as follow

...

Code Block
ssh -L 6006:localhost:6006 -o ProxyCommand="ssh -W %h:%p izar.epfl.ch" -l user i39

## with config
ssh -L 6006:localhost:6006 i39.izar

Image Modified

...