Sand glass topology: A simple and scalable topological structure for interconnection networks

Interconnection networks play an important role in parallel systems’ performance. They determine how processors are connected to provide high-performance and reliable connections between any two processors. A wide variety of interconnection network topologies have been proposed. This paper proposes a simple and scalable topology for interconnection networks called sand glass topology (SGT). The topological structure and topological properties of SGT are presented and analyzed. The topological structure of SGT combines line and ring topologies, which are simple and scalable. Adding more nodes or processors makes it easy to expand SGT to a larger size. A comparative study is conducted regarding various topological properties, including size, cost, connectivity, bisection width, and degree between SGT and several existing topologies, including full binary tree, non-wraparound mesh, and hypercube. The comparison aims to specify the rank of the proposed SGT among the other topologies. The comparative study concluded that SGT is a cost-effective topology that outperforms most aforementioned topologies. Moreover, it has been proved that SGT is a Hamiltonian graph with a Hamiltonian cycle.