A Reliability and Performance Model For Static Data Flow Machines

Abstract

The major advantages of parallel processing sys tems are their great reliability and high perfor mance. A class of massively parallel computing systems is the data flow machines. These machines work on the basis of data flow rather than control flow. This paper presents a reliability analysis of data flow machines using a graph theoretical ap proach. Three machines are considered here. They are the MIT, DDP and LAU static data flow ma chines. The data flow graph has been employed as a natural tool for representing that class of ma chines. The isomorphism between Petri nets and data flow graphs has been exploited to detect whether the consistency constraints are satisfied during various operational conditions. Such a graph is extended so that a timed data flow model has been constructed. This model integrates both the reliability features dependent on the system structure and the performance characteristics dependent on the components behavior. More over, a productivity index is introduced for evaluating the three machines.