Open OnDemand (web-based HPC access)¶

Overview¶

Open OnDemand provides a browser-based interface to UCT HPC.

It allows you to:

access HPC through a web browser
launch graphical applications without SSH configuration
run interactive sessions on compute nodes
manage files directly through a web interface

This represents a shift toward more accessible HPC workflows, reducing reliance on command-line interaction.

What Open OnDemand provides¶

Through Open OnDemand, you can:

launch interactive desktop sessions
run applications such as:
Jupyter notebooks
RStudio Server
MATLAB
VS Code (server-based)
open terminal sessions on compute nodes
browse, upload, and download files

All applications still run on HPC compute nodes, not on your local machine.

Key concept¶

Open OnDemand does not replace HPC scheduling.

Instead, it:

requests resources on your behalf
launches sessions on compute nodes
manages those sessions through the browser

This means:

you are still using the scheduler
resource limits still apply
sessions consume allocated compute time

When to use Open OnDemand¶

Use Open OnDemand when:

you need a graphical interface
you are new to HPC
you want a simpler way to run interactive workflows
you need persistent sessions you can reconnect to
you want to avoid SSH/X11 setup

Session behaviour¶

Open OnDemand sessions are:

interactive — you work in real time
stateful — you can disconnect and reconnect
resource-bound — limited by requested CPU, memory, and time

Always remember:

sessions consume HPC resources
long-running sessions should be used carefully

Relationship to other workflows¶

Open OnDemand complements, rather than replaces:

batch jobs (for large-scale computation)
command-line workflows (for automation and scripting)

Typical usage patterns:

exploratory analysis → Open OnDemand
production runs → batch jobs
debugging → interactive sessions

Limitations¶

Open OnDemand is not ideal for:

very large-scale parallel jobs
highly automated pipelines
workflows that require full control over scheduling

In these cases, use standard batch job submission.

Good practice¶

use Open OnDemand for interactive and exploratory work
avoid leaving sessions running unnecessarily
match requested resources to actual needs
move stable workflows into batch jobs

When to ask for help¶

Contact HPC support if:

an application you need is not available
sessions fail to start
performance is consistently poor
you are unsure how to map your workflow to Open OnDemand

Graphical applications → Reference > HPC
Scheduler and job submission → Reference > HPC
Software and modules → Reference > HPC
Submit your first job → How-to > HPC