Author(s)

A.J. Davies

Abstract

Massively parallel computing and the boundary element method A.J. Davies Dzr252072 o/ AW/temaZzca, [/mrerszZ?/ o/ Hertfordshire, Hatfield, UK ABSTRACT Massively parallel computing sytems belong to the classification SIMD, Single Instruction Multiple Data [1]. In such systems a very large number of relatively unsophisticated processors are connected together in some fashion and there is very rapid transfer of data between adjacent processors. Each processor receives the same instruction but works on its own data. Each processor performs a relatively simple task and for this reason the SIMD machine has a fine-grained parallelism. The boundary element method comprises three distinct phases: (i) equation set-up, (ii) equation solution, (hi) recovery of the field variables. All three phases exhibit a fine-grained parallelism [2], [3], [4], [5] and this parallelism has a straightforward mapping onto an SIMD architecture In this paper we shall consider

Keywords