[not urgent] IPython notebooks running on Hal and accessed locally

[kuod]

There's some interest in running IPython notebooks on Hal and accessing them locally.

This solution suggests starting the notebook on a compute node then creating a single SSH tunnel from the local machine with credentials to access first the head node and then the compute node.

An alternative method uses two different SSH tunnels as follows:

• create the IPython notebook from an interactive compute node
• SSH from the headnode to the compute node
• SSH from my local machine to the headnode

My question is this: what is the preferred method of tunneling between my local machine, the head node and the compute node to access the IPython notebook session?

[tatarsky]

I am on vacation so I will let folks mull this and if I see problems with the combined wisdom I will chime in when I return.

The main concern I would have is items running on nodes should be via the scheduler so we do not have untracked resources. But otherwise this has not come up to my knowledge on Hal to have a preference yet.

Paul Tatarsky paul@clusterguys.com

On Aug 17, 2015, at 6:00 PM, David notifications@github.com wrote:

There's some interest in running IPython notebooks on Hal and accessing them locally.

This solution suggests starting the notebook on a compute node then creating a single SSH tunnel from the local machine with credentials to access first the head node and then the compute node.

An alternative method uses two different SSH tunnels as follows:

create the IPython notebook from an interactive compute node SSH from the headnode to the compute node SSH from my local machine to the headnode My question is this: what is the preferred method of tunneling between my local machine, the head node and the compute node to access the IPython notebook session?

— Reply to this email directly or view it on GitHub.

[tatarsky]

I see no collective wisdom on my return has been gathered ;)

Therefore there is no preference beyond my comment on if its on the nodes it needs to be via a scheduled job. If I've not grasped your question on preferences feel free to comment further.

[kuod]

@vipints is putting together a set of scripts that will enable this feature. I'll let them update once things are up and running, but we will heed your comment that nothing will fall out of the watchful eye of the scheduler.

[tatarsky]

OK. Basically if the walltime limits become an issue I've seen queues designed for long running but low resource items as well. So keep me in the loop but I'll wait to hear how it works out.

[tatarsky]

I assume this is good to go and no further input needed. If incorrect, feel free to reopen.