SESYNC’s Slurm compute cluster allows users to run big memory- and processor-intensive jobs. Many users don’t know that you can access the memory and processing power of the cluster interactively, typing commands directly into the command line or into an R or Python session. This FAQ briefly describes how to start an interactive job on the Slurm cluster.
First, you need to open a terminal window and connect to the SSH server. On Windows machines, if you want to use applications with a graphical interface (such as displaying R plots in real time), you will need to use a third-party client such as PuTTY or MobaXTerm. See this FAQ on accessing Linux resources for help on getting the client set up.
Log into the SSH server by typing
ssh (your username goes here)@ssh.sesync.org
into the terminal prompt, and then entering your password when prompted. You are now logged in to the server called
sshgw00 (SSH gateway).
Next, start an interactive job. This is done with the shell command
srun, which requests a node for a parallel job. A typical interactive job request is:
srun -n 1 --pty bash -i
-n 1 means you are requesting one node. The
--pty bash -i part tells the job to open an interactive terminal. You can add other options such as requesting specific amounts of memory or time. See the Slurm documentation page for srun or the SESYNC cluster Quickstart page for more information — the available options for
srun are mostly the same as for
If the node(s) you requested are available you will get another terminal prompt with the name of the Slurm node you are logged into. Your interactive job is now running!
Now that your interactive job has started, you can do things like start an R session from the command line. First you will need to load the
Rcommon environment module, which loads R and all commonly used dependencies such as GDAL:
module load Rcommon
You can also load specific versions of R — see the environment modules Quickstart page for more information.
Now start the R session. This is as easy as typing
R but it is often preferable to type
That will ensure that your R workspace will not automatically save to your working directory and clutter things up in case your job terminates abruptly.
You can now run R code from the command line, including code that runs in parallel if you want. If you run code to draw and view a plot, the display will be forwarded from the remote machine to a window on your local machine. (This can be sluggish sometimes so it is often better to write the plot to a file and view it later.)
Once you are done with your job, quit R using
q() (or quit whatever other application you are running). Then type
into the command prompt. Your terminal prompt will say
sshgw00 again, meaning you are back on the SSH gateway and the nodes you ran your job on are now freed up for other users.