Table des matières

Running Jupyter notebook with V100 and SGE

We can run Jupyter notebook on the Volta nodes using SGE and ssh tunneling. This is done in 3 steps:

  1. Submit SGE Job(login session)
  2. Start Jupyter notebook in the opened shell
  3. Establish ssh tunnel and connect to the notebook

Submit SGE Job

Request one GPU and 40Gb of memory for 4 hours :

$ qlogin -q volta.q -l h_vmem=40G -l h_rt=4:00:00

Start Jupyter Notebook

In the opened shell session, we need to start jupyter-notbook. We will use conda to isolate a new environment, for example.

But first, as usual we need to modify HOME directory to WORK (because HOME is Read Only)

$ export HOME=$WORK

We load anaconda and conda:

$ module load anaconda

We create or connect to our isolated conda environment. Suppose that we have already an environment (use conda env list to view your environment list):

$ conda activate /Work/Users/kmazouzi/conda/meso

Finally we start jupyter-notebook inside:

$ (/Work/Users/kmazouzi/conda/meso) kmazouzi@node2-69:~$ jupyter-notebook --no-browser --ip=0.0.0.0

commands options are :

Connect to the notebook

If all goes well we see this message with important information

[I 11:04:54.299 NotebookApp] JupyterLab extension loaded from /opt/anaconda3/lib/python3.7/site-packages/jupyterlab
[I 11:04:54.299 NotebookApp] JupyterLab application directory is /opt/anaconda3/share/jupyter/lab
[I 11:04:54.300 NotebookApp] Serving notebooks from local directory: /Work/Users/kmazouzi
[I 11:04:54.300 NotebookApp] The Jupyter Notebook is running at:
[I 11:04:54.301 NotebookApp] http://(node2-69 or 127.0.0.1):8888/?token=07f2f688b651f329387c7b1459389edc7d1f37da60c90d65
[I 11:04:54.301 NotebookApp] Use Control-C to stop this server and shut down all kernels (twice to skip confirmation).
[C 11:04:54.307 NotebookApp] 
 
    To access the notebook, open this file in a browser:
        file:///Work/Users/kmazouzi/.local/share/jupyter/runtime/nbserver-2683-open.html
    Or copy and paste one of these URLs:
        http://(node2-69 or 127.0.0.1):8888/?token=07f2f688b651f329387c7b1459389edc7d1f37da60c90d65

There are three important pieces of information, above, to remember:

We can't connect directly to the link above because node2-69 is located in private network. We have to use ssh port tunneling.

SSH tunnel

We need to establish a tunnel from mesoshared to this volta node. Here is the command to use from mesoshared terminal:

$ ssh -g  -N -L 9999:localhost:8888 kmazouzi@node2-69

This command will map remote port 8888 from node2-69 to local port 9999 in mesoshared (localhost). Finally we will modify the URL above to connect to our Jupyter session.

OLD URL proposed by Jupyter:

The new URL to use, from your browser: