Zoom Introduction to the course & cluster access

For this course, we will use our training cluster UU (readers of Terry Pratchett may smile).

For interactive jobs requiring little computing power, we will use a JupyterHub running on UU and using few resources. We could run such jobs with salloc from an SSH session, but JupyterLab creates a friendly interface.
This is a good place to develop and try code, explore data, etc.

When we need lots of computing power however—for instance while training models—we will SSH to the training cluster and run sbatch jobs.
Training models from Jupyter would be a poor use of computing resources: while large amounts of memory, many CPUs and a GPU are useful to train models, having them allocated for an entire Jupyter session would have them sit idle for most of the time. This is something to think about when you use the real Compute Canada clusters!

JupyterLab

To launch JupyterLab on UU:

1. Go to https://uu.c3.ca
2. Sign in with your username and password
3. Set the server options (see image below):
     Please leave the following untouched:
       Reservation: "None"
       Account: "def-sponsor00"
       Number of cores: "1"
       GPU configuration: "None"
     These are the only values you should edit:
       Change the time to 8.0
       Change the memory to 2000
       Make sure the interface is set to JupyterLab
4. Press start

Note that, unlike other JupyterHubs you might have used (e.g. syzygy.ca), this JupyterHub is not permanent, so you can only use it for this course.

$ ssh userxxx@uu.c3.ca
# enter password (it is blind typing, so you won't see it as you type)