Welcome to Self-Management Use of HPC!
Getting Started on Colibrí
This guide is designed to support all users of the Colibrí HPC cluster—from interactive graphical workflows through Open OnDemand to automated pipelines via Slurm. Three core web-based platforms unify administrative management, computational access, and performance insights into a streamlined portal environment.

ColdFront
Request projects, join allocations, and manage research team roles before launching jobs.

Open OnDemand
Access Jupyter, RStudio, desktop GUIs, and file managers straight from your web browser.

XDMoD
Analyze core-hour burn rates, job efficiency, and memory metrics to refine workflow sizing.
The cluster integrates four primary platform technologies:
1. Accounts, Projects, and Allocations (ColdFront)
1.1 What ColdFront Is Used For
ColdFront is the official portal and system of record for:
- Requesting and renewing research projects
- Requesting CPU/GPU compute allocations
- Managing user roles and memberships within a project
- Tracking allocation consumption (also visible in XDMoD)
Note: You must belong to an active ColdFront project allocation to run jobs or launch interactive web sessions.
1.2 Common User Roles
- Principal Investigator (PI): Owns the project, submits initial requests, and manages allocations.
- Project Manager: Assists the PI with managing group users and allocation renewals.
- User: Executes computing workflows using the group's assigned resources.
2. Interactive Computing (Open OnDemand)
Open OnDemand (OOD) provides a fully web-based portal to Colibrí. No SSH keys, command line familiarity, or local client installs are needed to run interactive graphical applications.
2.1 Key OnDemand Features
- Interactive Apps: One-click launch setup for JupyterLab, RStudio, MATLAB, and XFCE Linux Desktops.
- File Explorer: Web-based file management—upload, download, edit, and move dataset files on PanFS.
- Web Shell: Full terminal access inside your browser window without needing client SSH setup.
- Job Composer: Graphical interface to draft, edit, and launch Slurm submission scripts visually.
3. Monitoring Usage & Performance (XDMoD)
XDMoD (Open XDMetrics on Demand) provides telemetry graphs and performance metrics to help users and PIs monitor cluster efficiency.
3.1 What You Can Analytics Track in XDMoD
- Resource Burn Rates: Track exact CPU and GPU hours used across projects over time.
- Job Efficiency Diagnostics: Compare requested memory vs. actual peak memory usage to prevent over-allocation.
- Queue Health: Analyze job wait times across cluster partitions to find optimal submission windows.
4. Choosing the Right Workflow Path
| Your Research Requirement | Recommended Workflow Path | Primary Tools Involved |
|---|---|---|
| Interactive GUI & Data Analysis | Open OnDemand Interactive Apps | OnDemand + PanFS |
| Workshops & Classroom Teaching | Open OnDemand Web Desktop | OnDemand Browser |
| Automated Pipelines & Batch Production | Slurm Batch Scripts (sbatch) |
Slurm + PanFS |
| Large Multi-Node MPI Jobs | Command Line Slurm Scheduler | Slurm SSH |
| Tracking Allocation & Efficiency Metrics | Web Telemetry Dashboards | ColdFront + XDMoD |
If you have allocation questions, access issues, or performance optimization needs, reach out to the Research Computing team via the HPC Support Portal or check your project status in ColdFront.