Tetrahedral Hyper-Cube: A New Scalable Hybrid Interconnection Network for Massive Parallel Processing

Abstract

This paper introduces an enhanced highly scalable, recursive and hierarchical interconnection topology called the Tetrahedral Hyper Cube (THC) for high end computing systems. As compared to the other interconnection networks THC is found to be more attractive in terms of various topological attributes such as node connectivity, edges, average node distance and message traffic density etc. The proposed network is bipartite and highly scalable and also robust in nature   than the other popular network topologies like Hyper cube (HC), Folded Hypercube (FHC), Exchanged hyper cube (EC), Mobius cube (MC) and Torcube (TC). With increased node count, the average node distance and message traffic density for THC is relatively small. The main attractive property of THC is its comparatively low link complexity which makes it a superior choice for massive scale computing. The optimal peer to peer minimal routing and broadcasting algorithms for the new network are also presented. The isoefficiency of THC lies between 0.6 to 0.56 for a varied magnitude of dimension. The packing density is four times that of the n dimensional hypercube. The detail analysis reveals that the THC topology is an efficient enhancement of the popular Hypercube topology and its variants.