The composite binary cube — a family of interconnection networks for multiprocessors

A class of interconnection networks called composite binary n-cube networks is presented in this paper for multiprocessors. These networks provide a spectrum of performance levels (measured in terms of message delays or network bandwidth) at different switching costs. At the low end of this spectrum is a structure called the expanded indirect binary n-cube (a multistage network), and at the high end is the well known hypercube structure. All members of this class have the same external connectivity between nodes, but utilize different internal switching architectures within the nodes. The distributed control algorithm for these networks is similar to those for multistage and hypercube architectures. An analytical model for the performance of these networks is also presented in this paper. Results of this model show that message delays decrease by about 40% from the low end architecture to the high end of the spectrum; however, the improvements in performance diminish near the high end. Thus for best performance/cost ratios within this class, our studies indicate that an intermediate architecture needs to be chosen.