Parallel Programs Development

摘要

A useful model for parallel computation should have the following properties: architecture independence, intellectual abstractness, software development methodology, cost measures, no preferred scale of granularity, efficiently implementable. This paper presents a model based on one developed by K.M. Chandy and S. Taylor, which is improved by defining cost measures and some methods for correctness proving (especially related to streams); and the reasoning for a program is made starting from postconditions. There are presented examples from Numerical Analysis - Symmetric System Solving, which reveal the power of the model. The algorithm used for Triangular System Solving is based on an efficient algorithm, presented by R.H.Biesseling, and have few modifications.