A Hypercube-based Scalable Interconnection Network for Massively Parallel Computing

Abstract

An important issues in the design of interconnection networks for massively parallel computers is scalability. A new scalable interconnection network topology, called Double-Loop Hypercube (DLH), is proposed. The DLH network combines the positive features of the hypercube topology, such as small diameter, high connectivity, symmetry and simple routing, and the scalability and constant node degree of a new double-loop topology. The DLH network can maintain a constant node degree regardless of the increase in the network size. The nodes of the DLH network adopt the hybrid coding combining Johnson code and Gray code. The hybrid coding scheme can make routing algorithms simple and efficient. Both unicasting and broadcasting routing algorithms are designed for the DLH network, and it is based on the hybrid coding scheme. A detailed analysis shows that the DLH network is a better interconnection network in the properties of topology and the performance of communication. Moreover, it also adopts a three-dimensional optical design methodology based on free-space optics. The optical implementation has totally space-invariant connection patterns at every node, which enables the DLH to be highly amenable to optical implementation using simple and efficient large space-bandwidth product space-invariant optical elements.