How to use Jupyter notebooks on the GPU nodes

Introduction

Users can run jupyter notebooks directly on the GPU nodes (or any other node) while accessing the notebeook from a browser within a ThinLinc session, or via an ssh tunnel.

Overview

start an interactive session
load an anaconda module
start a jupyter notebook in the background
access from a web browser in
- in a ThinLinc remote desktop session
- over an ssh tunnel
terminate the juptyer notebook when done

Full Procedure

Request an interactive session on a GPU node
```
 srun --account=my-account-id --partition=gpu --gpus=1 --time=2:00:00 --x11 --pty bash
```
When the interactive session is allocated, this will provide you with a command prompt on the allocated node e.g.
```
[acs03XXX@gpu01]$
```

load an anaconda module

module load anaconda/python-3.8.8/2021.05

start a jupyter notebook Note: make sure you use the name of the allocated node above (gpu01 in this example)
```
jupyter-notebook --no-browser --ip=gpu01  > notebook-output  2>&1 &
```
This command will send all output (and errors) from the juptyter-notebook process to the file 'notebook-output'. The final "&" places the process in the background.
find the session token for your notebook by typing:
```
jupyter notebook list
```
This will return something like:
```
http://gpu01:8888/?token=4f1f3e780d70d75c09b4ce6c77289f52cdd79047bb461665 :: /users/acs03XXX
```
The above information should also be visible in the 'notebook-output' file.
On the login node (from within a ThinLinc session) start a web broswer, and go to the URL displayed above:
```
http://gpu01:8888/?token=4f1f3e780d70d75c09b4ce6c77289f52cdd79047bb461665
```
Terminate the jupyter notebook by clicking on "Quit" in the web-browser, or by typing:
```
jupyter notebook stop
```