A roadmap to exascale agent-based modelling

Organizers:

Matt Hare (The James Hutton Institute, UK)

Gary Polhill (The James Hutton Institute, UK)

Doug Salt (The James Hutton Institute, UK)

Ricardo Colasanti (University of Glasgow)

Alison Heppenstall (University of Glasgow)

Richard Milton (University College London)

Michael Batty (University College London)

Description:

Exascale computers are capable of 10^18 floating point operations per second – the equivalent power of a billion laptops. This scale of computational power is achieved primarily through massive parallelism, in particular, though not exclusively, through GPU programming. There are embarrassingly parallel problems that modellers of complex, coupled social-environmental systems can trivially address through parallel computing architectures, not least in calibration over multidimensional parameter spaces. However, large-scale computing architectures such as exascale computers also provide opportunities to model systems in greater detail and with larger spatial extent, albeit that interactions among system components typically put upper bounds on the performance improvements that can be realized through parallelism. In the ‘ExAMPLER’ project, funded by the UK’s Engineering and Physical Sciences Research Council, we have been exploring how access to exascale computing could transform the use of agent-based modelling in policy-relevant contexts. This workshop builds on earlier visioning work to develop insights into how this transformative change could be realized through reimagined software architectures and institutions supporting high-performance computing use. The project has a website at https://exascale.hutton.ac.uk/”