I teach CSCI 356 (introduction to computer science) every semester, with about 120 to 200 students.
In the past, we’ve used a cloud instance (e.g., 32 CPU cores, 64 GB of RAM) every semester using department budget or cloud credits (all the students share the same instance).
Does CARC provide computing resources for classes?
That would be very useful! I’ve asked to carc-support@usc.edu but haven’t received any reply.
Hi,
I responded to your ticket but it looks like the email must have gotten mixed up. In regards to your request, we offer class allocations for compute time and 1TB storage free.
You can create a new project specific for your course and give it a storage and compute allocation. In the drop down menu you should see a specific option for courses.
If you have TAs, we also recommend that you designate them as a “manager” for the project so they can make changes if needed. Let me know if you have any other questions.
Thank you! We usually allocate on shared x86-64 Linux server and create SSH accounts for the students. Would we be able to do something like this with class allocations?
We use the Slurm job scheduler to manage access to compute resources. Users request compute resources for a specific amount of time and when the job completes, the compute node is added back into the pool of available resources for the next job to begin.
In this model your students can ssh to one of our login nodes and request an interactive session for a few hours while they work on their assginment. The filesystem is shared across the whole cluster so as long as they save their work, they will be able to start the process over again in a new interactive job.
If they don’t want to use the command line interface, we also have a few different graphical apps available through Open OnDemand, which is a web interface to the cluster.
https://www.carc.usc.edu/user-guides/carc-ondemand
You can also see our guide on using the job scheduler here:
https://www.carc.usc.edu/user-guides/hpc-systems/using-our-hpc-systems/running-jobs.html
So, if I understand correctly, students may experience queueing of their slurm requests, competing with each other and other users for access? And it’s not possible to reserve an instance?
Yes, they will have to wait in queue but most jobs start pretty quickly. If you had 100-200 students sharing 1 server, then they would probably only need 1 cpu and a few GB. Small jobs like that usually start within a minute or two.
If you want, we can set up a reservation of compute nodes during your class time so you and your class can run demos.