Evaluation of cube and data manipulator networks

The interconnection of a large number of processors and other devices to form a parallel/distributed computing system is a research area receiving a great deal of attention. One method is to use a multistage network. This paper compares two classes of multistage networks by examining two representative networks: the Generalized Cube and the Augmented Data Manipulator. The two topologies are compared using a graph model. By interpreting the graphical representations of the networks in different ways, different but functionally equivalent implementations result. The costs of the various implementations are compared taking VLSI considerations into account. Finally, the robustness (fault tolerance) of the different networks is measured and contrasted.