A model for dataflow computations: result sharing and its performance evaluation

While the dataflow model of computation is a popular way to model concurrent processing it ignores the potential for fast and efficient problem solving which is possible with the reuse of past computational results (result sharing). The concurrent processing and result sharing aspects in problem solving have been integrated into a unified scheme called concurrent processing with result sharing (CPRS) model (1987). In the current paper the CPRS model and the notion of CPRS decomposition based on problem dynamics are presented. The feasibility of the model is demonstrated by developing a dynamic dataflow architecture called the CPRS architecture. A simulation study of this architecture is used to evaluate its performance for solving realistic applications