Description
Discover how Research Infrastructure Engineers (RIEs) and Research Software Engineers (RSEs) can work together to boost HPC user confidence and improve cluster efficiency. We’ll be sharing results from the first ACIT hub hackathon and getting under the hood of the SLURM scheduler. The session concludes with an interactive unconference, inviting participants to reflect on limitations in current systems and explore what future schedulers, tools and teams can and should do better in an evolving landscape of data‑intensive and AI-driven research.
Dean Roe - Reflections on a Performance Monitoring Hackathon
What happens when you give a room full of HPC savvy RIE's and RSE's real cluster data, modern monitoring tools, and a single day to build something useful? This talk shares insights from an ACIT Hub hackathon exploring GPU performance monitoring, where teams rapidly turned ideas into working dashboards. Rather than focusing on tools alone, we reflect on what we learnt and what surprised us along the way. The session offers a glimpse into how hands-on, collaborative formats can surface challenges, change perspectives, and spark new approaches to managing accelerated compute.
Dean is the Head of Research Computing at the University of Surrey. Drawing on over 10 years of experience in Linux System administration, Dean has successfully developed a team of skilled HPC specialists, supporting team members through degree apprenticeship programmes and mentoring schemes. He is a champion for the adoption of DevOps practices in research environments.
Tina Friedrich - Demystifying the scheduler (SLURM)
SLURM underpins most HPC systems, yet its behaviour often appears opaque to users. This talk demystifies how the scheduler makes decisions, focusing on priority, fairshare, backfilling, and resource constraints, and why these can produce counterintuitive outcomes such as long queue times despite idle resources. We also explore key limitations, including reliance on user-supplied estimates and difficulty handling complex workflows. The session aims to move beyond mechanics to provoke discussion about what schedulers can and cannot do, and what future systems should look like in an evolving landscape of data-intensive and AI-driven research.
Tina Friedrich is a Senior HPC Systems Administrator in the Advanced Research Computing (ARC) team at the University of Oxford, which provides access to High Performance Computing resources, support, and advice to researchers within the University of Oxford. Before joining Oxford, they worked at Diamond Light Source, managing HPC systems and contributing to various systems administration responsibilities. They hold a Master’s in Physics from the University of Heidelberg.