Instructor information

We highly encourage all instructors to show demos and offer hands-on exercises on HPC systems, whether Compute Canada production clusters or simulated clusters inside a VM. Please prepare your setup well before the school begins: it is your own responsibility to make sure that in the class all attendees can log in and follow exercises on the cluster of your choice, and that this cluster has the necessary setup and software.

Cedar and Graham

Guest accounts

For this summer school, we have prepared 120 guest accounts and reservations on Cedar and Graham. On these two clusters, we prefer all attendees to use their own Compute Canada accounts during the school, however, some will not have an account, hence guest accounts. We will be providing guest accounts on small slips of paper before the beginning of your class.

Reservations

For the summer school, we have created Slurm reservations on Cedar and Graham. To run jobs under these reservations, please use (and pass to your audience) the following Slurm flags:

--account=ucss19-wa_cpu --reservation=ucss19-wr_cpu

Please note that these reservations will start on Sunday (May-26) morning, one day early for testing. All real account holders will need to be added to the reservation. We will try to add all registrants with confirmed accounts before the start of the school, but since we are working with partial information, we might need to add some attendees on the fly during your course. We will be coordinating this before your class.

Please note that Compute Canada production clusters can be down for emergency maintenance work (this has happened with us in the past with little previous warning), the scheduler or one of the parallel file systems can be overwhelmed with too many requests, etc. In other words, a production system is not guaranteed to be available for demos and training 24/7, so make sure you have a backup.

Training cluster in an Arbutus VM

Instructors are free to set their own VMs emulating a Compute Canada cluster on Arbutus system. These VMs are limited by your cloud account quotas, and Compute Canada / WestGrid staff have the same quotas as all other users and cannot spin arbitrarily large VMs. A typical training VM would have 5-8 nodes, two virtual CPUs and 3GB memory per node. Therefore, we discourage sharing these VMs between concurrent courses, as irresponsible use in one session can render this VM unusable in the other. Instead, we encourage instructors to set and maintain their own VMs which they can do with a Compute Canada account.

A training VM has many advantages:

  1. You can set up a very large number of guest accounts, and you can reset the passwords at any time.
  2. This training cluster will have Compute Canada’s CVMFS software stack mounted, available via the same module environment as on our general-purpose clusters Cedar and Graham. Not all software will work, though, due to licensing limitations (Intel compiler) and hardware differences. Slurm, GNU compiler and a lot of other CVMFS-installed software will work.
  3. You completely control the software stack and you can install anything as root.
  4. You completely control the scheduler, can kill any offending jobs, can restart the scheduler, etc.

Please note that these training clusters does not have GPUs. If you need to run demos and/or hands-on exercises on GPUs, you can use the production systems Cedar and Graham.

If you want to set up your own cluster, please contact alex dot razoumov at westgrid dot ca well before the school.

Course requirements

All courses must have hands-on exercises for the attendees. Whenever possible, courses must show examples in a proper HPC environment. For example, when demonstrating a workflow in a programming language, show the entire process (writing code, compiling it, and running it in a Slurm job) on the cluster, not just inside the IDE on your laptop. If applicable to your course, please demonstrate scaling up to large problem sizes and many cores/nodes.

If you need any help in porting your workflow to an HPC system, please let the summer school organizers know early on.

Please send to us

  1. your approximate course syllabus (you can always iterate on it later),
  2. any information you would like to pass to attendees: what software they need to install on their laptop, prerequisites for this course, the level of difficulty (beginner / intermediate / expert), any external links, and
  3. if using significant cluster resources, what hardware configuration you need per each student, e.g., (2 MPI tasks) x (3 cores per task)

Please check the online description of your course in the program and let us know if you want to add or change anything.