The Distributed Pipeline

Due to the crossbar organization of typical multiprocessors, their cost grows as the square of the number of processors. The distributed pipeline (DP) is a concept which allows the construction and operation of very large multiprocessors; an example having 300 CPU´s is described. The DP is an extension of the more traditional pipeline which has been used to implement highly concurrent computers. The DP allows increased concurrency, throughput determined by the average processing time, and the ability to operate at reduced size when one of the elements fails. Each element in a DP can implement a range of functions with little concern for speed matching between elements. The DP is therefore somewhat more general in application than the traditional pipeline, and has certain characteristics which are highly desirable in multiprocessors, e. g., potential for fault tolerance and simple throughput prediction.