On the attractiveness of the star network

In defiance of the tremendous research the star network had attracted during the last decade, no parallel machine based on this network has yet been built. In this paper, we explore the possible reasons that made the star network not so viable for actual implementation. First we show that the major deficiencies related to poor scalability, cumbersome routing, and unsuitability for real applications make the construction of star-based parallel machines not so appealing. Furthermore, we show that each of the star network variants solves some of the deficiencies, but fails to solve others and introduces new concerns. We also show that the star network and its variants will find it difficult to compete with simpler topologies such as binary cubes, meshes, and k-ary n-cubes in terms of simplicity and suitability for real applications. The results in this paper can be used in two ways. In one hand, they can be used to discourage further research that might end up useless, and on the other h...

• Type: Journal
• Year: 2002
• Where: JSS
• Authors: Abdel Elah Al-Ayyoub, Khaled Day, Abdullah Al-Dhelaan
• Keywords: JSS 2002, Parallel Machines, Real Applications, Star Network