ITS offers classroom use of high performance computing (HPC) resources on the Great Lakes High Performance Computing Cluster through Advanced Research Computing.
Details
ARC will cover the cost of access to Great Lakes at the rate of $60.91 per student, per semester. For multi-semester courses, contact ARC Support to receive the funding all at once.
- Ex. A class has 30 students for one semester at the rate of $60.91 per student. The course will receive an allocation of $1,827.30.
Funding will be provided in a lump sum to the class account. It is up to the instructor to allocate resources by individual or team.
To ensure HPC resources are available to classes for short-running jobs when the cluster is under heavy utilization, we have reservations set on the standard and GPU partitions that set aside nodes for use. The reservations will apply automatically to the following:
- Standard partition: jobs running for 2 hours or less
- GPU partition: jobs running for 1 hour or less
Default Limits
To promote fairness and ensure efficient use of resources for coursework, user time limits are set to encourage quicker turnover, allowing other students and cluster users access to the hardware.
Additionally, a default per-student limit will be placed on the account to avoid single-student runaway spending. If you are familiar with HPC Slurm Accounts, this differs from the norm in that the limit is set to each user instead of at the account level.
Most limits can be seen in the Resource Management Portal. Instructors can request default and single-user limits be changed at any time. This may be necessary if a single-user requests a job that will use more than the user’s remaining limits.
| Limit Type | Default Per User | Default Per Course | Squeue Reason Job is not Running |
|---|---|---|---|
| Spending Limit | $60.91 | $60.91 * # of students | AssocGrpBillingMinutes |
| GPUs in use | 1 | 5 | AssocGrpGRES |
| Running Jobs | 2 running (5000 queued) | n/a | |
| Maximum Duration/ Walltime | 8 hours | n/a | AssocMaxWallDurationPerJobLimit |
| CPU Limit | none | 500 | AssocGrpCpuLimit |
| Memory Limit | None | 7GBytes * CPU Limit | AssocGrpMemLimit |
Course instructors can determine how the entire course’s allocation is distributed across students at the single-user level. The allocations can be used to support individual, group, or major project work.
There can be restrictions on runtime, GPU, cores, and memory to reduce spending by students (the fewer resources one can request, the less a job will cost).
In the event the class computation requirements exhaust the allocation, another HPC Slurm Account can be created. Instructors will need to provide a shortcode to cover HPC costs after the no-cost academic resources have been used.
Exclusions: Courses that are not “classic courses” with instruction, syllabi, assignments, and grading cannot receive course support for HPC access.
Ex. Graduate thesis courses or courses that reflect faculty led research.
Getting Started
To request a course account, please contact ARC Support with the following information:
- Uniqnames of students to be put on the account
- Uniqnames of individuals to administer the account (i.e. instructor(s))
- Any limits to be placed on either the users or account as a whole
- The unit abbreviation and course and section numbers for the course that will receive the support (i.e. EECS498, Section 400)