Guest Editor's Introduction Interconnection Networks for Parallel and Distributed Processing

W ith the rapid advances in technol'VV ogy, it is now feasible to build a system consisting of hundreds or thousands of processors. Processors in such a parallel/distributed system may spend a considerable amount of time just communicating among themselves unless an efficient interconnection network (IN) connects them. A complete interconnection such as crossbar may be cost prohibitive, but a shared-bus interconnection may be inefficient and unreliable. We need to design INs with both cost and performance between these two extremes. Parallel or distributed computers can generally be divided into two categories: multiprocessors and multicomputers. The main difference between the two lies in the level at which interactions between the processors occur. A multiprocessor system must permit all processors to directly share main memory, as shown in Figure la. In a multicomputer system, however, each processor has its own local memory. Sharing between the processors occurs at a higher level, through a complete file or data set. As shown in Figure lb, a processor cannot directly access another processor's local memory. An IN is a complex connection of switches and links that permits data communication between processors and memories in a multiprocessor system or between the processors in a multicomputer system. A multiprocessor or a multicomputer architecture is further characterized by the topology of the IN it uses. Current multiprocessor organizations are based on crossbar, multistage interconnection networks (MINs) and multiple-bus networks, as shown by the first four articles in this special issue. Multicomputer architectures include topologies such as star, ring, tree, and hypercube, as shown by the last article in this issue.