Acceleration of Extreme Scale Flow Simulations through Hierarchical Mesh Partitioning

Abstract

This paper investigates how extreme scale computational fluid dynamics (CFD) and computational solid mechanics (CSM) simulations can be accelerated through hierarchical graph partitioning, i.e., partitioning a graph representing the unstructured mesh used for these numerical simulations on multiple hierarchy levels with respect to the architecture of high-performance computing (HPC) systems. Hierarchical partitioning can reduce communication between higher hardware hierarchy levels and, thus, reduce the communication time within simulations. This effect is expected to be greater with bigger meshes and a higher number of compute nodes used for these simulations. Consequently, applying hierarchical partitioning to industrial scale-resolving simulations can speed up and enable larger simulations of CFD, CSM, and coupled fluid-structure interaction (FSI) simulations than before. We propose an implementation of hierarchical partitioning and apply it to a very large test case with more than 1 billion cells.