Published 21/04/2026 By nAG
When selecting HPC systems the user interface is rarely a top priority for decision makers. Commonly higher on the priority list is system reliability, support contracts, faster time to first result, and an easier path to running software correctly. Yet these criteria fail to consider the capabilities and limitations of the systems users, often leading to a disconnect between them and the IT department that maintains them.
Open OnDemand is an open-source project the nAG HPC Services Team have integrated into existing HPC environments to provide a more accessible, graphical way for users to interact with complex systems 1. Out of the box, it provides a solid foundation, but typically requires configuration to align with existing infrastructure, including authentication services, single sign-on, and certificate management.
Beyond initial deployment, it can be adapted to reflect the organisation’s existing services, including branding, interface design, and user experience. It can also be extended with additional observability tooling, as well as application-specific workflows that reflect how users interact with the system in practice. When implemented effectively, this allows Open OnDemand to become a fully integrated layer within the HPC environment, helping to make day-to-day operation more seamless for both users and administrators.
In this blog we explore how well-designed GUIs can embed best practice into HPC workflows, aligning user behaviour with system policy while improving both usability and operational efficiency.
From a HPC sysadmin perspective, the highest-value GUI workflows are the ones that prevent avoidable cluster harm without turning the centre into the “no” department. Many centres impose policies on users, such as login nodes are shared and must not be used for computationally intensive work, and they direct users to submit through the scheduler instead 2. From our own experience, we find inexperienced users don’t always follow such guidance, which results in a poor user experience and additional burden on sysadmins to resolve. The training burden to avoid repeat failures then becomes an additional overhead for successful system operation.
Open OnDemand streamlines access to complex applications across engineering, simulation, and research domains. Whether working with CFD tools such as Ansys Fluent, structural analysis packages like Abaqus, or AI workloads such as AlphaFold, users can launch and manage jobs through a simple web interface. Integrated remote desktop and VDI capabilities further enhance this experience, enabling a seamless and efficient workflow from anywhere. This simplicity is driven by powerful, configurable controls within Open OnDemand, including defaults, policy enforcement, validation, templating, scheduler integration, license management, idle session handling, and observability.
Even with these capabilities in place, the underlying user behaviours remain consistent. Scientists are not trying to break the cluster. They are trying to finish a paper, a design iteration, or a grant milestone, often under tight time and resource constraints. The problem is that the default HPC interface is still largely SSH plus tribal knowledge, which is a combination that makes doing things such as “quick tests” on login nodes tempting.
Another common pattern is accidental resource wastage; be that through a forgotten job option or a missing walltime parameter. None of this comes from a place of malice. It is what humans do when the system expects them to be part-time schedulers.
By presenting users with a GUI each of these behaviours can be overridden, with defaults explicitly set or fields marked as required where necessary. The extra mental overhead imposed on the users immediately dissipates and their workflow becomes more efficient.
The real power of a GUI in HPC isn’t that it makes things easier. It’s that it makes the correct way of using the system the path of least resistance for both experienced and in-experienced users. A well-designed GUI reduced the burdens of on-boarding and skill-fade by embedding those policies directly into the submission path.
Instead of telling users not to run compute workloads on login nodes, the GUI simply doesn’t offer that as a pathway. Interactive applications are launched onto compute nodes by default. The “wrong” option disappears, not through enforcement after the fact, but by design.
Screenshot shows a bespoke Open OnDemand interface view, highlighting how the GUI can be tailored to suit user needs (see additional views below)
Similarly, scheduler complexity can be reshaped into structured input. Walltime becomes a required field with sensible defaults. Partitions are presented as a filtered dropdown based on what the user is actually allowed to access. Resource requests can be bounded, guided, or pre-filled based on the application being launched. In each case, the system is now asking the user to describe their problem in a controlled way instead of expecting the user to translate their intentions into scheduler syntax.
Importantly, this is not about restricting users. It is about encoding institutional knowledge into reusable templates, informed by how systems are actually used in practice. An Open OnDemand application is a defined set of instructions which depicts how it should be run in your environment. That includes the right modules, the correct scheduler flags, appropriate resource ranges, and any site-specific constraints.
In this model, policies stop being advisory and start being executable. And perhaps most importantly, users stop needing to think like schedulers in order to do their work.
OnDemand is highly configurable and can be customised to requirements, even after deployment, enabling organisations to iteratively adapt workflows, interfaces, and integrations as user needs evolve, without needing to fundamentally rework the platform.
From interfaces to run simple wrapper calls to web-based APIs, or even more complex middleware type tools, the support for custom applications in OnDemand is nearly limitless 3.
It is also strong in terms of scheduler support. Open OnDemand includes built-in support for both Slurm and PBS Pro (as well as several others 4), and allows you to customise submission and control commands through wrapper scripts. This is particularly useful in cases where a remote scheduler needs to be connected over SSH, for example.
For repeatability or container-based deployments using the official OSC Ansible role is recommended 7. The Ansible role is highly extensible and well-suited for automation. If for example a new version of OnDemand is released and verification that customisations function correctly before updating production is needed, Ansible provides an effective way to do so.
In short, a well-designed GUI can reshape how HPC systems are used. By embedding best practices directly into the workflow, it reduces waste and operating costs, protects shared data, other content resources, and accelerates user productivity. The result is a system where doing the right thing is effortless, and both users and administrators benefit, particularly when those workflows are carefully aligned to the needs of the organisation.
In practice, implementing these workflows effectively depends on how well system policy, user behaviour, and application requirements are brought together. If this is something you are exploring within your own environment, we are always happy to share experience or discuss approaches and share a live demo. Connect with the HPC Team to schedule a meeting.
Screenshots showcase different views of the Open OnDemand interface, highlighting how the GUI can be tailored to suit user needs – work undertaken by the nAG HPC Services team
