Optimising a High Performance Computing allocation scheme for maximum utilisation

Mr John Zaitseff1, Mr Martin Thompson1

1UNSW Sydney, Sydney, Australia

Like many other research institutions in Australia, UNSW Sydney provides the majority of its High Performance Computing (HPC) resources using a partner allocation scheme at the National Computational Infrastructure (NCI) facility in Canberra.

However, apportioning these HPC resources to various projects can be quite inefficient in terms of overall scheme usage, depending as it does on researchers making accurate forecasts of their usage patterns.  In fact, it is not unusual for such schemes to experience a utilisation of only 70–80%, representing an equivalent loss of 20–30% of research funds.

This talk shows how Research Technology Services at UNSW Sydney has significantly and consistently mitigated this problem through on-going micro-optimisations of the allocation scheme to achieve utilisation rates of close to 100% on average.  Such micro-optimisations allow us to keep in close contact with researchers, to better understand their computing requirements.


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.

https://orcid.org/0000-0003-4504-1124

Categories