Parallelization of a highly unstructured Euler-solver based on arbitrary polygonal control volumes

Publisher Summary This chapter discusses the Parallelization of a Highly Unstructured Euler-Solver Based on Arbitrary Polygonal Control Volumes. The Euler algorithm is a cell-centered finite volume method with higher order reconstruction in two and three space dimensions with explicit integration in time. The reconstruction is based on derivatives calculated from a locally fixed neighborhood and ensures positivity of the relevant unknown density, pressure, and internal energy. A flexible grid representation for finite volume methods is proposed. Polyhedra with an arbitrary number of sides are used as control volumes. The chapter describes a strategy to obtain single program multiple data (SPMD)-parallelization by using an abstract model for distributed, dynamic data. The abstract model encapsulates the distributed aspects of the application and provides an application-oriented interface to message passing. The strategy is applied to a self-adaptive finite volume scheme based on arbitrary polyhedral control volumes. Efficiency and numerical results are presented. The chapter diagrammatically presents parallel efficiency measured on an Intel Paragon system.