Performance of Multiprocessor Interconnection

w ith device characterist ics approaching physical limits, parallel or distributed process-ing has been widely advocated as a promis-ing approach for building high per-formance computing systems. The con-tinued impetus in research in these areas arises from two factors: (a) the technolog-ical development in the areaof VLSI chips and (b) the observation that significant exploitable software parallelism is inher-ent in many scientific and engineering applications. To exploit this parallelism efficiently, a parallel/distributed system must be designed to considerably reduce the com-munication overhead between the proces