SC22 Proceedings

The International Conference for High Performance Computing, Networking, Storage, and Analysis

Birds of a Feather Archive

Scientific Code Coupling on Supercomputers

Authors: Stephen Longshaw (Science and Technology Facilities Council (STFC), UK Research and Innovation), David Emerson (STFC Daresbury Laboratory, England; UK Research and Innovation), Garth Wells (University of Cambridge), Chris Richardson (University of Cambridge), Yu-Hang Tang (NVIDIA Corporation), Timo Betcke (University College London), Nick Brown (Edinburgh Parallel Computing Centre (EPCC)), Benedict Rogers (University of Manchester), Georgios Fourtakas (University of Manchester), Rupak Biswas (NASA), Lorena Barba (George Washington University)

Abstract: This session will explore the state-of-the-art in scientific code coupling, paying particular attention to enabling software currently in development. Consideration will be paid to the state of existing libraries in the context of exascale computing, mathematical rigour, and corresponding workflows, with highlighting examples of applications drawn from several scientific areas. We will look at whether the current trajectory of coupling technologies is the right one and if so, what can we do to improve core performance, portability, and applicability to enable massive, coupled simulations on supercomputers.

Long Description: The concept of multi-physics or multi-scale modelling is now ubiquitous across computational science. The capability of combining and utilising multiple methods or discretisation schemes is not just an enabler of better science, it’s often imperative to understand the physical challenges presented to modelling and simulation today. This session is designed to consider the now established topic of scientific code coupling with a focus on the software that enables it.

Key software libraries already exist that are designed to enable the coupling of scientific codes. Most are designed with high-performance computing in mind and offer similar capabilities in coupling terms. However, there are key differences between them and as we move to exascale and look to couple models with many billions of cells or particles, a key question is what these differences are and what is the best overall solution amongst them?

Use in high-performance computing environments, in particular heterogeneous supercomputers, is paramount and needs a deep discussion. The scope for library-level offload and acceleration (for example through GPU) of computational overhead introduced by coupling (such as spatial interpolation) is clear. The question is how to best achieve this and what technologies will provide the optimal platform.

Furthermore, generalised scientific code coupling is now an established topic, with a number of key software implementations available (e.g. MUI; preCICE; PLE; CWIPI; MUSCLE; ADIOS). Conversation is needed to ensure that the directions it takes next mean it becomes a key enabler for the next generation of scientific modelling and simulation. We often look to hero simulations from those few codes (and problems) that scale to justify the need for huge supercomputers, why not also look to combine the capabilities of many codes to achieve whole machine scaling at the exascale and beyond?

This BoF is relevant to a number of groups of people: those who actively produce new coupled scientific software solutions; those who develop or actively use code coupling technologies in their work; those interested in exploring how coupled (especially partitioned) solutions can address exascale-level computational challenges; those interested in the current state-of-the-art for scientific code coupling, especially within the context of supercomputing; those who consider the wider implications of massive coupled simulations or integration of Machine Learning models, such as data and model interoperability, reproducibility of software stacks and simulation results and code portability.

The aims of this BoF are to provide a forum for those involved in the topic of scientific code coupling to discuss and explore the area and define the state of the art; to offer a wider audience from the supercomputing community with a clear overview of this emerging and important topic and demonstrate its importance in the new exascale landscape; to help refine the goals of those actively working in this area with the aim of ensuring a clear and joined up approach to code coupling emerges over the coming years.


Back to Birds of a Feather Archive Listing