Instructor Guide
Classroom & Instructional Use Guide
This companion guide is for instructors using the HPC cluster for teaching, workshops, or training. It focuses on common instructional workflows and administrative decisions, rather than deep technical mechanics.
Need general user documentation? Direct your students to the main User Guide for cluster connection details, environment setup, and command syntax.
1. Before the Semester or Workshop
1.1 Requesting Allocations (ColdFront)
Submit a request for a course or workshop project in ColdFront well in advance. Ensure project names are clearly identifiable (e.g., PHYS-410-Spring-2026).
- Interactive Open OnDemand sessions
- Regular homework assignments and labs
- Final projects (peak resource utilization period)
1.2 Adding Students & Verification
Select the onboarding model that best matches your course structure:
- Self-Request Access: Recommended for large classes. Students apply to your ColdFront project ID directly.
- Manual Addition: Recommended for small, intensive workshops. Instructors upload student usernames.
2. Recommended Teaching Model
2.1 Default to Open OnDemand
For most instructional courses, Open OnDemand (OOD) should serve as the primary interface.
- Eliminates local OS setup and SSH learning overhead
- Performs reliably in classroom and remote environments
- Supports Jupyter Notebooks, RStudio, MATLAB, and web terminals
- Allows immediate success without prior Linux/HPC experience
2.2 When Slurm Is Appropriate
Introduce batch queueing via Slurm only if it directly aligns with course learning objectives, such as:
- Teaching reproducible computational research
- Demonstrating cluster scaling, memory usage, or queue concepts
- Executing batch parameter sweeps or multi-node analyses
Avoid enforcing Slurm usage purely as a gatekeeping prerequisite.
3. Structuring Assignments
Design course assignments around brief walltimes (minutes to 1 hour), low CPU core counts, and minimal memory allocations. This ensures rapid job turnaround and prevents queue congestion during peak submission hours.
Supply starter Slurm job templates, pre-configured Jupyter/R scripts, and standardized directory structures within PanFS. Consistency drastically minimizes basic setup errors and support tickets.
4. Data Management (PanFS)
- Home Directory: Best for source code, small scripts, and configuration files.
- Scratch/Project Space: Mandatory target for large computational output and training datasets.
- Local Laptops: Discourage storing primary results locally; keep workflows cluster-bound.
5. Monitoring Class Progress
5.1 Class Analytics via XDMoDInstructors can monitor metric dashboards in XDMoD to identify runaway jobs, spot inefficient CPU/memory requests, and track allocation burn rates before major assignment deadlines.
5.2 Explicit Syllabus ExpectationsDefine policy upfront regarding interactive vs. batch usage, resource availability limits, and submission protocols during peak queue periods.
6. Support & Classroom Strategies
- Practice Assignment: Schedule a non-graded "dry run" early in the term to catch authentication and environment setup issues.
- Support Protocols: Train students to include exact Slurm Job IDs and error messages when requesting help.
- HPC Coordination: Notify system support prior to class-wide project deadlines or live workshop sessions.
7. Common Pitfalls to Avoid
| Common Instructional Pitfall | Recommended Alternative Practice |
|---|---|
| Requiring SSH for absolute beginners | Use Open OnDemand for browser-based terminal access and GUI tools. |
| Unconstrained resource assignments | Set strict walltime limits and low core/memory defaults for lab exercises. |
| Unannounced enrollment changes | Update ColdFront project rosters promptly when students drop or add. |
| Assuming intuitive filesystem knowledge | Provide explicit directory templates and clear rules for home vs. scratch usage. |
Final Advice for Instructors
Successful instructional use of high-performance computing prioritizes:
- Accessibility over technical complexity
- Reproducibility over raw computational performance
- Clear expectations over implicit operational norms
When in doubt, simplify—and let the computing serve the learning goals, not the other way around.