Open OnDemand

This document describes how to use TJ's Open OnDemand instance

What is Open OnDemand?

Open OnDemand is how you can access TJ's Cluster. It contains many different applications within it, including JupyterLab, a fully-online IDE that is very similar to Google's Colab interface. Additionally, it also contains other IDEs for other programming languages, including a fully online VS Code via Codeserver! There are also plans to add more IDEs, such as RStudio and/or MATLAB in the near future.

How do I open an Application?

TJ's OnDemand instance is available at https://ondemand.tjhsst.edu. The first time you load the page, it may take a minute or two, and you will be redirected to Ion to authenticate. Then you will be presented with a screen that looks like this:

Here, choose the application you want. In the following, we assume you choose Jupyter. You should then be asked for a Job Profile, with the following options:

• Default (4 CPUs, 8GB RAM)

• Compute (8 CPUs, 16GB RAM)

• 1 GPU (8 CPUs, 16 GB RAM, 1 GPU)

• 2 GPUs (8 CPUs, 16 GB RAM, 2 GPUs)

• 4 GPUs (8 CPUs, 16 GB RAM, 4 GPUs)

Please only choose what you need. Depending on usage, it may take longer to spawn instances for higher resource allocations, and others may be prevented from using resources allocated to you if you are not using them.

You will then be placed in a queue to wait for your own server to spawn. Our instance is configured such that your server may be spawned on one of 53 different machines. Even though your server is spawned on random machines, all your files are stored in a common location: /csl/users/<username>.

The page will automatically update and show you the following screen once it has loaded:

At this point, click on the blue "Connect to Jupyter" button to start the Jupyter Lab session.

Advanced Features

Python Interfaces for JupyterLab

After logging in, you may be greeted with a "Launcher" tab. This tab will have multiple options, including:

• Python Notebook

• Python Console

• Terminal

The Python Notebook and Python Console are two separate interfaces, and you may experiment with either of them.

As you see in the top two rows, the options "Notebook" and "Console" are available. Notice how they are both named "Python (default)"

Both of these Python interfaces are based in the "default" Python environment, meaning they have the default packages. In most cases the default packages will suffice, but if you need to install custom packages read the Open OnDemand sections.

Activating Ready-made Environments

The (amazing) Sysadmins provide everyone with a set of predefined environments on the launcher for different purposes (e.g. Physics, Quantum Computing, Computer Vision, etc). To activate these environments while using the terminal, you must run the following command:

module load <name of environment>

Note that <name of environment> should be replaced with the name of an environment. For example, to activate the physics environment, run:

module load physics

To see a list of all available modules, run:

module avail                       # see all available modules
module show <name of environment>  # more details information about a specific module

To undo the loading of a module, simply run:

module unload <name of environment>

Creating Custom Pixi Environments

Both the Python Notebook and Console are backed by Pixi environments. Familiarize yourself with Pixi environments before reading on. If you're familiar with conda environments, Pixi is a faster conda with more features.

To create a personal custom Pixi environment, do the following: on infoprism or any cluster node run the following in your home directory (/csl/users/<your_csl_username>)

pixi init my_personal_environment
cd my_personal_environent
pixi add ipykernel "python>=3.13"

my_personal_environment can be replaced with whatever you want.

After creating the Pixi environment, activate it:

pixi shell

Finally, add the environment to JupyterLab:

python3 -m ipykernel install --user --name "my_env" --display-name "My Env"

The --name and --display-name parameters can be anything you wish.

After running the previous commands, your launcher should now look like this:

Notice how a second option for both Notebook and Console is now available. The names match the "--display-name" parameter in the previous step.

Now you have created a personal environment you have total control of. You may use pixi to install any custom packages you wish. Be sure to click on your new Python Notebook/Console option (not the default option) if you wish to access your custom environment.

Removing Environments

Removing environments is a permanent action and cannot be reversed!

Removing the Pixi environment is simple:

rm -rf /csl/users/<username>/my_personal_environment

Removing the entry is a little more involved:

rm -r /csl/users/<username>/.local/share/jupyter/kernels/<name>

• The <name> parameter must match the --name parameter passed to the python -m ipykernel ... step, NOT --display-name! You may also ls around in this directory to find your matching environment and then rm -r the directory

• After removing the directory, reload JupyterHub and the entry will now be removed.