Issues in designing truly scalable interconnection networks

Interconnection networks have been extensively studied in the computer architecture community as they are essential to the performance of parallel processing computing platforms. However, this research area has also been criticized for producing many practically useless research results, which hinder the needed progress in this area. This paper begins with a new classification of interconnection networks and then discusses major practical issues in the design of interconnection networks. The emphasis will be on "truly scalable" interconnection networks. A more practical definition of "incremental scalability" is elaborated. Just like the evolution of microprocessors, as technology advances, the architecture of interconnection networks may also change. Understanding those technological constraints and performance demands is important to future research in this area.