Astrohack

Team Brussels Commuters

Set up jupyter notebook on a supercomputer over a double tunnel.

First, pick a 4-digit number unique to you (8888 in this example, but pick something else)

From your local machine:

ssh -i id_vcs [email protected] -L 8888:localhost:8888

(or may god help you with setting up port forwarding with putty)

On the login node

Set up a conda environment containing jupyter (for example using one of the environments from the astrohack repo but adding jupyter in the .ymlfile, or conda install jupyter in the environment)

Request your compute note if needed.

module load tmux
tmux

# For CPU:
qsub -I -A lp_astrohack -lnodes=1:ppn=20 -lwalltime=4:00:00
# For GPU:
qsub -I -A lp_astrohack -lpartition=gpu -lnodes=1:ppn=20:K40c -lwalltime=4:00:00

<CTRL+SPACE> d

Find its name with

qstat -n

Assuming the name is r1i0n1 we can set up the second tunnel with

module load tmux
tmux
ssh r1i0n1 -L 7333:localhost:7333
# Now you're on the compute node!
cd $VSC_DATA
source activate keras
jupyter notebook --port=7333
<CTRL+SPACE> d

and copy-paste the link in the first lines of the output to go to your notebook server.