Optimization of the Communications between Processors in a General Parallel Computing Approach Using the Selected Data Technique

摘要

A large variety of problems that are out of reach of single processor computer capabilities. Many approaches are offered today to get round this. Each of these has its own strengths and weaknesses : a compromise has to be found. We will introduce a general parallel computing method for engineering problems dedicated to all users. We have searched an easy method for code development. A technique of data selection (Selected Data Technique - SDT) is used for the determination of the data dedicated to each processor. Several problems associated with the communication times are posed and solutions are proposed in accordance with the number of processors. This method is applied to very large CPU cost problems, particularly the unsteady problems or steady problems using an iterative method. So the domain of potential applications is very wide. The SDT-parallelization is performed by an expert system called AMS (Automatic Multi-grid System) included in the software. This new concept is a natural way for the standardization of parallel codes. An example is presented hereafter.