Preparing For Multidisciplinary Science In The Exascale Era

Abstract:

Exascale computing has many dimensions to it, and as anticipated, it has many more than the predecessor steps involved with either terascale or petascale. The changes needed to address more complex science include significant value and opportunity to many non-traditional HPC communities. While not denying the value of broad access to information and associated standards for very distributed activities, the voracious needs of our peak science cases for performant, scalable, flexible and well-connected infrastructure means that co-location and strongly coordinated activities are needed, while allowing for external access via appropriate services and information systems.

A question for HPC centres as they prepare for 2030, is what types of services should be provided by an exascale facility. How should we manage the various needs of multidisciplinary science through both co-design, and harmonisation across individual communities, and to support multi-disciplinary communities.

At a technical level, the changes range from adapting major codes to fundamental hardware changes, improvements in efficiencies and methods of codes while maintaining reproducibility, the management of a wider diversity of software, the heavy reliance on both access to, and management of data, new forms of scalable processing. The improvements of infrastructure are now more continuous and layered. Much more attention is placed on what it means to test and evaluate quality of service, and how to transition in a robust and manageable way for downstream users or additional layers of dependent infrastructure.

In this talk I will discuss a number of aspects to this, including successes and use-cases from several scientific communities.