Properties and Performance of Linearly Extensible Multiprocessor Interconnection Networks

Abstract

Interconnection networks arise as the basic bone of research related to parallel computing, high-performance system design, distributed computing or computer networks. There has been a strong interest in the design of modern interconnection topologies to achieve the desired performance. This paper presents the performance study of a new class of parallel architectures known as linearly extensible multiprocessor architectures. It describes the properties and parameters to evaluate the performance of newly designed interconnection networks. The comparative study is carried out which shows that the proposed class posses the desirable topological properties of hypercube as well as tree type topologies. The performance of these architectures is also evaluated in terms of load balancing by applying standard scheduling algorithm on them. The simulation results show that the proposed architectures are performing on equal footing and their performance are compatible with standard hypercube architecture. The comparative study implies the various aspects while designing and efficient multiprocessor interconnection network.