Pruning Generalized Hypercube Interconnection Networks for Diameter Preservation: RedCube

摘要

Interconnection Networks are key components of parallel systems, networks on chips and data center switches. They may be employed to connect cores, processors, processors and memories, or computers. The topologies of interconnection networks in the literature constitute a very wide spectrum. One major topology for such networks is the Generalized Hypercube. It has simple routing and low distances between its nodes but a very high number of links per node (degree of node). This means a high number of I/O ports. In this paper, a “pruning” algorithm is presented that reduces the number of links per node in Generalized Hypercubes while virtually maintaining the distances between nodes. In some identified and specified cases, the network diameter is maintained. The network diameter is the maximum value among the set of shortest distances between all node pairs in a network. The resulting reduced network, the RedCube, constitutes a class of topologies that has low distances, low diameters and simple routing. Basic comparisons show the advantages of the RedCube over other common network topologies.