Running High Performance Computing as a Service to enable research at scale

Mr John Zaitseff¹, Mr Martin Thompson¹

¹UNSW Sydney, Sydney, Australia

High Performance Computing (HPC) is recognised as a strategic research enabler for universities.  Providing appropriate HPC facilities to enable research at scale is, however, a significant challenge, requiring substantial investment and commitment.  This talk covers how UNSW Sydney has risen to this challenge by providing HPC as a Service for researchers at the University.

Research Technology Services (ResTech) at UNSW was formed in November 2017 to provide a “step change” in how HPC is offered.  Instead of continuing to support individual aging faculty- or group-based HPC clusters, a strategic decision to outsource the majority of HPC resources allowed UNSW to rapidly expand computationally intensive research by an order of magnitude.  These resources are sourced from a variety of external partners including those overseas.

On-going micro-optimisation of the resource allocation scheme allows the University to consistently achieve utilisation rates of close to 100% on average.  This prevents a loss of 20–30% of research funds on resource underutilisation.  More importantly, it allows us to keep in close contact with researchers, to understand their requirements.

In addition to these partner schemes, ResTech maintains a UNSW-wide HPC facility, Katana, that provides a safe, friendly environment for researchers beginning their HPC journey, enables flexible and bespoke computational needs not provided by partners or cloud resources, and encourages co-investment using research grants by offering surrounding infrastructure support including networking, hardware, software and security support.

Such strategies have been successful in enabling research requiring high performance computing at a scale previously unseen at UNSW.

Biography:

John Zaitseff is a Research Computing Support Engineer at UNSW Sydney.  He provides both wide and deep expertise in all aspects of High Performance and Research Computing to researchers, including analysing computing needs and requirements, providing advice on purchasing and utilisation, training staff and students to use new and existing facilities, system design and administration, and performing software design, programming, debugging and code optimisation using multiple programming languages.