NI-HPC+: Increased High Performance Computing for Data Science Applications

The £5M NI-HPC Tier 2 national supercomputing centre (https://www.ni-hpc.ac.uk/) and the Kelvin2 cluster were established in November 2019 with support of a £2.1 million grant from EPSRC (EP/T022175). Overall, 35% of the processing time is dedicated to supporting general users for the national Tier-2 service, with the remaining resource allocated to research projects led by the host universities. It is the first national HPC and AI centre in Northern Ireland and a collaboration between QUB and UU. It was at the forefront on the adoption of lower power AMD CPU technology matched with a high percentage of Graphics Processing Unit (GPU) rich capability with suitably high bandwidth to support data-driven artificial intelligence research.

The Kelvin-2 facility went into production on target in April 2020 with research exemplar users given early access and opening to the wider community in late summer 2020. Operational service levels have had a satisfactory level of 99.9% availability. The original growth plans to expand the HPC user base are continuing to track ahead of target, seeing 'active user' numbers increase by 90% in the last 16 months; utilization of the cluster has reached 80% for compute nodes. The users have been supported by a dedicated team including two RSEs, funded under the EP/T022175 grant. They joined the team in July 2020, delivering training and user support to increase cluster utilization and manage queue times.

The proposal is to build upon this success and respond to the increased need by increasing the capacity of the NI-HPC site by 25%. To achieve this, £0.5M of funding is being sought to provide 16 low power, AMD CPU units each with 128 cores and 1TB of memory, and 2 GPU nodes each with 4 * A100 units. The current projection on usage will see the existing cluster usage saturating by summer/autumn 2022, so this call is a timely opportunity to invest in the successful growth of NI-HPC. The plans to increase capability for the EPSRC HPC user base are focused on expanding the research around the six key exemplars (R1-R6) and three new areas R7-R9. In addition to the exemplar groups, NI-HPC supported the effort to fight the COVID-19 pandemic, an activity that was highlighted on the University webpage.

An area of improvement has been the slow take-up by UK researchers outside Northern Ireland. A number of initiatives are in hand to address this including hosting the NI-HPC user conference and RSE champions event in Belfast, and engaging an experienced Network Manager to promote and publicise the facility. The NI-HPC team at its regular quarterly exemplar team meetings continues to challenge themselves to on-board UK based collaborators.

We continue to deliver on building the UK's skills base and software development community required for the continued operation of high quality HPC services and the introduction of modern industry approaches to Service Management. Evidence includes the Gold Star Award from the Service Desk Institute for the process and teams in support of NI-HPC and the wider Queens IT estate. The NI-HPC team is actively engaged with the RSE Champions' network and are organizing the October 2021 event.

This proposal includes a detailed implementation plan and clear focus on risk management addressing equipment delivery, HPC Service Delivery and Cyber Security. Queens Information Services has already ordered compute and GPU equipment for HPC usage in the 21/22 academic year. Queens are happy to divert this equipment to the NI-HPC work and order replacement delivery to mitigate delivery risk and re-order new equipment for its needs as Queens financial year ends 31st July. NI-HPC is the one of the first Tier 2 sites to engage with JISC on Cyber Essentials accreditation to mitigate the cyber risk.