Alan Aitken: Fast, scalable, computations of subglacial sediment processes for improved Ice-Sheet models

Alan Aitken: Fast, scalable, computations of subglacial sediment processes for improved Ice-Sheet models

The authors have developed the model code GraphSSeT to compute sediment transport and beneath ice sheets and glaciers. GraphSSeT provides the first comprehensive approach to modelling subglacial sediment system and these provide the ability to resolve boundary conditions as well as grain size and detrital provenance to inform and constrain models of ice sheet evolution and sea level rise. Previous generations of GraphSSeT showed slow compute performance relative to hydrology codes and poor scaling in distributed compute environments, limiting the opportunities for broad application. This presentation covers the progress made under the Pawsey PULSE collaborative project to improve performance. The redesigned codebase focuses on fast and scalable parallel code with GPU support enabled. The project focused on improving task scheduling and reducing message-passing overheads as well as changes to underlying data structures and functions, but with minimal change to the user-experience. Some examples of the new version’s performance are shown for small medium and large model cases.