A new era of accessible high-performance computing at UCT

The recent upgrade of the University of Cape Town’s (UCT) Upper Campus Data Centre has provided UCT with a world-class environment for housing the high-performance computing (HPC) facility. “With state-of-the-art cooling, power redundancy, and enhanced security, we’re able to ensure the reliability and performance that researchers at UCT require for cutting-edge computational work," said Tim Carr, senior technical specialist with UCT HPC.

Key to these developments, is the availability of OnDemand, a powerful new graphical interface designed to make HPC more accessible to the university's researchers. Traditionally, accessing UCT HPC resources required familiarity with command-line tools and Linux environments, which posed a barrier for many users. 

OnDemand addresses this by providing an intuitive, web-based interface that streamlines the process of submitting and managing computational jobs on the university’s advanced cluster, and opens up UCT’s advanced computational resources to a far broader range of disciplines and expertise levels.

UCT HPC: Offering secure, scalable, and container-ready research computing

At its core, UCT HPC is built on robust, open-source technologies. The platform leverages SLURM, the global standard for HPC workload scheduling, ensuring efficient management of computational jobs and optimal use of cluster resources. Authentication is seamlessly integrated with UCT’s Active Directory, allowing researchers to use their existing institutional credentials to access the cluster after approval by their faculty’s principal investigator. This workflow not only streamlines access but also ensures compliance and accountability.

Researchers can submit parallel and multi-threaded workloads and monitor job progress through a web-based interface that requires minimal technical background. The platform supports a range of computational needs, from data-intensive simulations to GPU-accelerated tasks, and offers robust storage options to ensure data integrity and high availability during research workflows.

One of the standout features of UCT HPC is its support for containerised applications using Singularity. Unlike Docker, which is not suitable for multi-user HPC environments due to security concerns, Singularity enables researchers to package their applications (including all dependencies) into portable containers. These containers can be built on a researcher’s laptop, another cluster, or even in the cloud - and then executed on the UCT HPC infrastructure without modification. 

This approach ensures both reproducibility and ease of software deployment, allowing researchers to update or modify their computational environments in a non-intrusive, layered manner. Singularity also supports the use of MPI (Message Passing Interface) for parallel workloads and integrates seamlessly with the cluster’s powerful NVIDIA A-100 GPUs, which can be partitioned to suit the needs of different projects.

The underlying storage infrastructure has also been significantly enhanced. UCT HPC utilises Pure Storage’s parallel NFS, allowing for high-throughput data operations and non-disruptive upgrades. This means that researchers no longer face downtime during maintenance. Three distinct storage volumes support different aspects of research workflows:

• /scratch for high-speed, temporary data;
• /home for scripts and configuration files (with backups); and
• /opt/exp_soft for institution-wide software deployments. 

Researchers are encouraged to move critical data to UCT’s Research Data Store (RDS) for long-term, backed-up storage, ensuring data integrity and compliance with university, South African and funder policies.

Democratising HPC access to drive innovation across disciplines

UCT eResearch is committed to maximising the adoption of open-source software, reducing costs, and avoiding vendor lock-in. OnDemand is designed to emulate the look and feel of commercial cloud services, providing researchers with a familiar experience while maintaining the advantages of on-premises infrastructure, such as data sovereignty, cost control, and tailored support.

“With the launch of OnDemand, we’re making high-performance computing at UCT truly accessible for the first time,” said Professor Mattia Vaccari, director of eResearch at UCT. “Researchers no longer need to be Linux experts or command-line wizards to harness the power of our HPC facility. This is a major step towards democratising advanced computing for all disciplines – from the humanities to health sciences – and ensuring that every researcher has the tools they need to innovate and make an impact.”

Pooling resources into the central facility offers greater efficiency and compliance, as maintaining individual machines in departmental spaces is less cost-effective, more prone to disruption (network and power interruptions), and fails to meet ISO 27001 security standards required by funders. 

The UCT HPC website contains information for new users including a ‘cheat sheet’ as well as more in-depth documentation on how to accomplish supporting tasks such as moving data or installing additional libraries. 

“HPC and Linux can be intimidating for first time researchers”, explained Andrew Lewis, senior technical specialist with UCT HPC. “We hold regular courses throughout the year on both introductory and advanced topics.”

Looking ahead, the team is exploring the potential for hybrid cloud integration and further enhancements to support evolving research needs. By lowering barriers-to-entry and continuously introducing new features, OnDemand is poised to accelerate scientific discovery at UCT, fostering innovation and empowering researchers across all fields to leverage the full potential of advanced computing.

Researchers interested in leveraging or investing in UCT HPC are encouraged to contact the eResearch Centre.