The Jupyter Notebook is a web-based interactive computing platform. The notebook combines live code, equations, narrative text, visualizations, interactive dashboards, and other media. There are a few ways to launch an interactive session to run a Jupyter notebook on Strelka, though the preferred way is to use conda.
Conda
Using conda, it is possible to create a dedicated, personalized environment in which to run Jupyter Notebooks. The following example creates a basic conda environment within which to run a Jupyter Notebook or Jupyter Lab application (more specific documentation for conda is available), and it assumes you are connected to Strelka and are running an X server to use graphical applications; note also that you should not launch Jupyter applications from the head node directly and instead should start an interactive session on a compute node before doing so:
conda create --name jupyter python=3.11 conda activate jupyter conda install jupyter jupyter-notebook ## launch a Jupyter Notebook jupyter-lab ## launch Jupyter Lab
Running either jupyter-notebook or jupyter-lab will open a web browser from Strelka and connect you to the Jupyter session. To exit the session, close the web browser and hit Ctrl-c to shutdown the server. Note that graphical performance is likely better using an interactive desktop through Open OnDemand.
This slightly more complex example demonstrates creating a conda environment with Jupyter and PyTorch, then connecting to a GPU node, requesting 4 CPUs, 32GB of memory, a single GPU, and then launching a notebook:
conda create --name jupyter-pytorch python=3.11 conda activate jupyter-pytorch conda install jupyter conda install pytorch salloc -t 300 --cpus-per-task=4 --mem-per-cpu=8gb --gpus-per-node=1 --x11 jupyter-notebook
Open OnDemand
For basic needs, it is also possible to launch an interactive Jupyter notebook session directly from Open OnDemand; the Jupyter session will run with 4 CPU cores and 1 GPU.
Create and connect to an interactive session
When using the interactive Jupyter notebook, your code will be running on Strelka worker nodes.
- Go to strelka.swarthmore.edu and log in with your Swarthmore credentials
- Click on Interactive Apps → Jupyter Notebook
- Set the number of hours you require. The maximum is 5 hours. After the time limit expires, the session will end.
- Click Launch
- Wait for the interactive job to be ready. If there are resources available, this will take about a minute. If the server is busy, other jobs on Strelka will need to complete before there are resources available for your Jupyter session.
- Click on the Connect to Jupyter button
If you are have a time-critical need or are planning for a group to use interactive Jupyter notebooks at the same time, get in touch with ITS at support@swarthmore.edu so we can plan for how make the resources available.
Switching to JupyterLab
The default view is Jupyter notebook. To switch to the JupyterLab interface, edit the URL in your browser: replace the /tree? at end of the url with /lab. JupyterLab is a newer interface that contains the notebook functionality along with other tools.
Jupyter Help Documentation
Jupyter Notebook: https://jupyter-notebook.readthedocs.io/en/latest/
JupyterLab: https://jupyterlab.readthedocs.io/en/latest/index.html
