Parallelization of a High-Order Accurate Unstructured Mesh Finite-Volume Solver

摘要

A generic solver eliminates the need to write new finite-volume codes for each type of physics.By separating the physics from the numerics of the solver, a modular design is achieved.New physics modules can easily be written with a minimal knowledge of the finite-volume method.A parallel solver allows simulation of complex physics on intricate domains in a timely manner by using numerous processors simultaneously.In this paper we describe the steps needed to adapt a high-order accurate unstructured mesh generic finite-volume solver to a parallel architecture.A message-passing approach is used which allows the solver to operate on a distributed memory system,such as a cluster of workstations.The re- construction stencil is determined at the preprocess- ing stage and an appropriate parallel data structure for the solution is formed.Fluxes for faces on the par- tition boundary are evaluated by communicating the reconstruction coefficients to the adjacent processor. Good performance scalability is achieved for second and fourth-order accurate solutions on cell and vertex centered meshes.