Performance of multiprocessor interconnection networks

L.N. Bhuyan, Qing Yang, D.P. Agrawal
1989 Computer  
w ith device characteristics approaching physical limits, parallel or distributed processing has been widely advocated as a promising approach for building high performance computing systems. The continued impetus in research in these areas arises from two factors: (a) the technological development in the areaof VLSI chips and (b) the observation that significant exploitable software parallelism is inherent in many scientific and engineering applications. To exploit this parallelism
more » ... a parallel/distributed system must be designed to considerably reduce the communication overhead between the processors. The communication architecture of the system might support one application well but might prove inefficient for others. Therefore, we need to take a general approach, independent of the application, while designing the communication system or the interconnection network (IN) of a general-purpose parallel/distributed system. The IN must be efficient, reliable, and cost effective. A complete interconnection, such as a crossbar, might be cost prohibitive, but a shared-bus interconnection might be inefficient and unreliable. Thus, present research is directed to designing INS whose cost and performance lie somewhere between the two extremes. Multiprocessor designers need analytical techniques to evaluate network performance. This article presents a tutorial on these evaluation tools to guide designers through the design process. Ongoing research in the area of parallel and distributed processing suggests a number of promising INS. Because of the high cost involved in hardware implementation or software simulation of these INS, performance evaluation of these networks needs t o be carried out through analytical techniques so that we can make a choice between various alternatives. A mathematical model makes it possible t o study the efficiency of the IN in terms of various design parameters used as inputs to a model. Therefore, the intent of this article is t o provide a tutorial on the subject of performance evaluation of multiprocessor interconnection networks to guide system designers in their design process. A classification of parallel/distributed systems. We can divide general-purpose parallel/distributed computer systems into two categories : multiprocessors and multicomputers. The main difference between them lies in the level at which interactions between the processors occur. A multiprocessor must permit all processors to directly share the main memory. All the processors address a common main memory space. In a multicomputer, however, each processor has its own memory space, and sharing between the processors occurs at a higher level as with a complete file or data set. A processor cannot directly access another processor's local memory. Multiprocessors can be further divided as tightlycoupled and loosely coupled. In a tightly coupled system, the main memory is situated at a central location so that the access time from any processor t o the February 1989 001 8-9l62/89/02oO-oO25S0l .WO I989 IEEE 25
doi:10.1109/2.19830 fatcat:5fp6yvp5frc4tg5f3aaz4v5ake