Performance Analysis of Multiprocessor Architecture
Hesham El‐Rewini, Mostafa Abd‐El‐Barr · 2004
In this Chapter, we provide an in-depth analysis of the performance measures of parallel architectures. Our coverage in this chapter starts by introducing the concept of computational models as related to multiprocessors. The emphasis here is on the computational aspects of the processing elements (processors). Two computational models are studied, namely the equal duration processes and the parallel computation with serial sections models. In studying these models, we discuss two measures. These are the speedup factor and the efficiency. The impact of the communication overhead on the overall speed performance of multiprocessors is emphasized in these models. Having introduced the computational models, we then present a number of arguments in support of parallel architectures. Following that, we study a number of performance measures (metrics) of interconnection networks. We define performance metrics such as the bandwidth, worst-case delay, utilization, average distance traveled by a message, cost, and interconnectivity. We show how to compute those measures for sample dynamic and static networks. Our coverage continues with a discussion on the scalability of parallel systems. A discussion on the important topic of benchmark performance concludes our coverage in this chapter.