Multiprocessor architectures for solving PDEs

For the past few years, there has been an increasing interest in the design and development of multiprocessor systems for solving very large scientific computing problems - in particular the solution of Partial Differential Equations (PDEs). This paper first presents (i) the motivation for building multiprocessor systems for solving PDEs, (ii) the key issues in the design of such multiprocessor systems, and (iii) a taxonomy of these systems. Several architectures of multiprocessor systems which have been specifically designed for solving PDEs are then discussed with special emphasis on an architecture proposed recently by the authors. A comparative study of these architectures is also included.