Parallel Computation on Hypercube-Like Machines.

The hypercube interconnection network has been recognized to be very suitable for a parallel computing architecture due to its attractive topological properties. Recently, several modified hypercubes have been propose to improve the performance of a hypercube. This dissertation deals with two modified hypercubes, the X-hypercube and the Z-cube. The X-hypercube is a variant of the hypercube, with the same amount of hardware but a diameter of only $\lceil$(n + 1)/2$\rceil$ in a hypercube of dimension n. The Z-cube has only 75 percent of the edges of a hypercube with the same number vertices and the same diameter as the hypercube. In this dissertation, we investigate some topological properties and the effectiveness of the X-hypercube and the Z-cube in their combinatorial and computational aspects. We give the optimal or nearly optimal data communication algorithms including routing, broadcasting, and census function for the X-hypercube and the Z-cube. We also give the optimal embedding algorithms between the X-hypercube and the hypercube. It is shown that the average distance between vertices in a X-hypercube is roughly 13/16 of that in a hypercube. This implies that a X-hypercube achieves the better average communication performance than a hypercube. In addition, a set of fundamental SIMD algorithms for a X-hypercube is given. Our results indicate that the X-hypercube makes an improvement in performance over the hypercube, but not as much as the reduction in a diameter, and the Z-cube is a good alternative for the hypercube as far as the VLSI implementation is of major concern.