Mesh-hypercube : a network topology for parallel systems

Abstract EN

Direct networks are popular way of interconnecting large number of processing elements of today’s large commercial and research multicomputers.

The design of such networks is very important because it has a large effect on the performance, cost, and the scalability of the system.

In this research we propose a new direct interconnection network suitable for large-scale parallel systems called the Mesh-Hypercube (MH).

The MH uses the hypercube topology and the mesh topology as its basic structure.

The MH combines the positive features of both the hypercube and the mesh, while at the same time circumventing their disadvantages.

Hypercube topology has small diameter, high connectivity, symmetry, simple routing, and fault-tolerance.

Mesh topology has simple regular structure, constant node degree, easily implemented, and highly size scalable.

A very attractive feature of the MH network is its ability to support a large number of processors while maintaining constant node degree.

In addition, the MH exhibits small diameter, small average distance, high connectivity, and simple routing.

Other positive features include incremental scalability and fault tolerance.