PISCES MP - Adaptation of a Dusty Deck for Multiprocessing

Abstract : Scalable multiprocessors offer high performance with relatively low cost. Unfortunately, the programming model required to take advantage of these architectures is a radical departure from traditional paradigms. Most users are unwilling to discard the knowledge and expertise captured by existing dusty-dock programs in exchange for a faster yet unproved and unfamiliar parallel code. To explore the potential for providing vastly improved dusty-dock performance while preserving the knowledge implicit in the program, we have parallelized the device simulator PISCES on an Intel iPSC/86OTM hypercube. Section II gives a brief overview of PISCES. Section III describes the methods used to transform PISCES into a parallel code. A demonstration of the computational power of the new parallel device solver is presented in Section IV. Improvements to the linear solver are discussed in Section V. Finally, conclusions are given in Section VI.