Jupyter Access

Running Jupyter Notebooks¶

Jupyter notebooks are great for interactive development. Before following these steps, ensure you have a working Miniconda/Python environment following the Getting Started Guide.

First, install JupyterLab in your environment:

# Activate your environment
conda activate myproject

# Install JupyterLab and useful extensions
pip install jupyterlab ipywidgets matplotlib

# Install kernel for this environment
python -m ipykernel install --user --name myproject --display-name "Python (myproject)"

Step 1: Request a GPU node with resources¶

srun --gres=gpu:1 --time=12:00:00 -p pi_ppliang --nodelist=node2500 \
     -c4 --mem=32G --pty bash

Adjust the resources based on your needs. Jupyter itself doesn't need much, but your computations might.

Step 2: Activate your environment¶

conda activate myproject

Step 3: Start JupyterLab on the compute node¶

# Choose a port number between 1024-65535 (e.g., 8888, 9999, 1717)
jupyter-lab --ip=0.0.0.0 --port=1717 --no-browser

The output will show something like:

[I 2024-01-15 10:30:00.123 ServerApp] Jupyter Server is running at:
[I 2024-01-15 10:30:00.123 ServerApp] http://node2500:1717/lab?token=abc123def456...
[I 2024-01-15 10:30:00.123 ServerApp] http://127.0.0.1:1717/lab?token=abc123def456...

Keep this terminal open! Copy the token from the URL - you'll need it.

Step 4: Create SSH tunnel (on your local machine)¶

Open a new terminal on your local computer and run:

ssh -N -L 1717:node2500:1717 <mit_username>@orcd-login001.mit.edu

This creates a tunnel: - -N: No command execution, just forwarding - -L 1717:node2500:1717: Forward local port 1717 to node2500 port 1717

Note

This terminal will appear to hang - that's normal! It's maintaining the tunnel.

Step 5: Access Jupyter in your browser¶

Open your web browser and navigate to:

http://127.0.0.1:1717/lab?token=abc123def456...

Use the complete URL with token from Step 3.

Step 6: Clean up when done¶

Always clean up to free resources:

Save your work in Jupyter
In Jupyter: File → Shut Down
On compute node terminal: Press Ctrl+C to stop JupyterLab
On local terminal: Press Ctrl+C to stop the SSH tunnel
Type exit on compute node to release the resources