An expert system approach to data distribution and distribution analysis

Abstract

Research into systems aimed at transforming existing legacy programs to an equivalent form for improved execution performance on multiprocessor systems has led to the realisation that the level of parallelization expertise needed is similar to that of a human expert. Data distribution is one of the major obstacles limiting pure automation in parallelization. A system which automatically provides effective data partitioning algorithms can be considered suitable in replacing the human expert [1]. The Fortport project provides a solution to this problem by presenting a suite of tools to apply AI technology in the parallelization of sequential codes. This approach is based on an underlying knowledge model to influence the transformation process.