Hands-on Workshop September 21, 2023

Please see the Clusters Guide and other documentation sections on the left for more information.

Contact: Megan Duff and Jan Mandel will be happy to answer any questions!

Log in

know your CU Denver username
open a Terminal (mac or linux), or command/powershell window (windows 10 or 11)
type into the window: ssh your_username@math-alderaan.ucdenver.pvt (replace your_username by your own username, of course)
Enter your CU Denver password

The easiest text editor is nano, exit it by control-x
mkdir to make a directory, cd to change directory. cd .. will go to the parent directory.
man commandname to get help about a command. Try man top

Look at the file alderaan_simple.sh
Submit it: sbatch alderaan_simple.sh
Look for the output file, it will have name starting with slurm
Try a GPU job: sbatch alderaan_single_gpu.sh What did it do? Look at the output.

Multiple cores within a single node - just put the number on the --ntasks or -n line other than 1 when using software that written to use multiple cores automatically, such as R or Matlab.
Multiple cores on multiple nodes - we have 2048 cores total on the compute nodes! - you need to have a code specifically written for this. Let's build some. Type cd examples make
Look at alderaan_single.sh alderaan_mpi_general.sh and submit them
Look at run_alderaan.slurm and submit it. What is it doing?

Complete computing environments with custom software and different Linux versions.
Our containers are in /storage/singularity. See Singularity for what containers we have and more details.
alderaan_single_gpu.sh you used before runs tensorflow in a singularity container. Look at the script how it works!
singularity_alderaan_shell.slurm allows you to run an entire shell script in a singularity container. Try to add another command. Try to use another container.
Extra credit: run the examples from https://github.com/ResearchComputing/Intro_GPU_Acceleration.

Another way to set up a custom environment is by modules. Type module avail what is there and man module for more information.

Use one of the singularity containers with anacoda and make your own conda envirohments.
Or install your own anaconda or miniconda.

Please do not ssh to work on compute nodes, you could interfere with jobs running there which would make you very unpopular. It is OK to ssh to compute nodes to check on your running jobs submitted through sbatch, however.
The magical incantation srun -p math-alderaan --time=2:00:0 -n 1 --pty bash -i will teleport your session to a compute node for two hours with one core reserved for you. Try it! Your interactive job will not interfere with CPU usage of other. Try matlab -nodesktop, run something CPU intensive, ssh to the node from another terminal window, and try top
Try a Python, R, or Matlab job!
Sorry no graphics

Our compute nodes have 64 cores and 512GB memory each. This seems like a lot, but you may be sharing it with others. We do not control memory as an allocatable resource yet, so if you need to use a large amount of memory, talk to us first.