Key Points

What is OnDemand?

  • OnDemand is a web-based interface for accessing Sagehen HPC resources
  • It requires no SSH knowledge and works in any modern web browser
  • OnDemand is ideal for file management, interactive computing, and job monitoring
  • Pomona’s OnDemand is available at https://ondemand.hpc.pomona.edu/
  • OnDemand complements SSH access rather than replacing it

Logging In and Dashboard Overview

  • Log in at https://ondemand.hpc.pomona.edu/ with Pomona credentials and DUO MFA
  • The top navigation bar has the logo, menu, help links, and account dropdown
  • The side menu organizes access through Files, Jobs, Apps, and Clusters
  • Color-coded indicators show job and resource status at a glance
  • Always log out on shared computers to protect your account

Navigating Storage Locations

  • Sagehen has four storage locations: /rhome, /bigdata, /scratch, and /tmpfs
  • /rhome and /bigdata are persistent and share a 1 TB lab quota on BeeGFS
  • /scratch is non-persistent SSD storage deleted when the job completes
  • /tmpfs is RAM-backed storage deleted when the job completes
  • Use quota_check.sh to check usage – du does not work correctly on BeeGFS
  • Choose storage based on persistence needs, speed requirements, and sharing

Uploading, Downloading, and Editing Files

  • Upload files via drag-and-drop, single select, or multi-select
  • Download single files directly or multiple files as a ZIP archive
  • Create, rename, move, and delete files through the graphical interface
  • Edit text files directly in the browser with syntax highlighting
  • File permissions can be viewed and changed through the web interface
  • Monitor your quota regularly with quota_check.sh or Accounts > Usage

Shell Access and Command-Line Computing

  • Web Terminal provides full command-line access through your browser
  • Launch from Interactive Apps or Clusters menu in OnDemand
  • Run standard Linux commands, submit SLURM jobs, and manage files
  • Multiple terminal sessions can run independently for parallel work
  • Use screen/tmux for long-running processes that survive disconnections
  • Web Terminal is convenient for beginners; SSH is available for advanced users

Launching Jupyter Notebooks

  • Launch Jupyter from Interactive Apps with configurable CPU, memory, and time
  • Expect 1–5 minutes for initial resource allocation
  • Use code cells for Python and markdown cells for documentation
  • Install packages with !pip install directly in notebook cells
  • Save work frequently and download notebooks before sessions expire
  • Use batch jobs for computations longer than 8 hours

RStudio and GPU Resources

  • Launch RStudio from Interactive Apps with configurable resources
  • RStudio provides script editor, console, environment, and files/plots panels
  • Install R packages from the console or the Packages tab
  • Request GPU resources by selecting the gpu partition and GPU type (A100, L40S, RTX PRO 6000)
  • Verify GPU access with torch.cuda.is_available() in Python or tf in R
  • Save your work frequently – sessions have time limits

Submitting Batch Jobs

  • Access the Job Composer through Jobs in the left sidebar
  • Quick Submit lets you paste a script; Upload lets you use an existing file
  • SLURM directives control partition, time, nodes, and output locations
  • Use –partition=gpu and –gres=gpu:1 for GPU jobs
  • Job arrays let you submit many similar jobs with one script
  • Sagehen HPC offers three partitions: amd, gpu (both with 30-day max walltime), and short (shorter max walltime, good for quick test/debug jobs — check sinfo -p short)

Monitoring Jobs and Troubleshooting

  • Active Jobs shows real-time status with color-coded indicators
  • Click any job to see details, output logs, and error logs
  • Common failures include wrong paths, missing modules, and insufficient resources
  • Cancel jobs from Active Jobs or with scancel in the terminal
  • Sagehen offers three partitions: amd and gpu (each up to 720 hours / 30 days) and short (shorter max walltime, good for quick test/debug jobs — check sinfo -p short)
  • Check cluster status before submitting and prefer off-peak hours for faster starts