Dataparallel implicit Navier–Stokes solver on different multiprocessors

Abstract

This chapter discusses the parallelization of a Navier–Stokes solver on the multiprocessors of different architectures and programming models, a system with shared memory, and several systems with distributed-memory. For communication, all multiprocessors use a message based model and are reproduced on the multiprocessor architecture by means of available communication models. This is done in such a way that only a machine specific interface has to be exchanged. The results obtained by the parallel Navier–Stokes solver on multiprocessor systems (transonic flow around a profile) are compared with those of workstations and super vector computers. The developed communication model has proofed as a flexible, reliable, and efficient implementation. High speed-ups are obtained on any multiprocessor because there are workstation clusters or massive parallel systems. It allows the computation of a wide range of block-structured grids and is easily transferable.