CONTROL OF BOTTLENECKS IN A NETWORK DATABASE MACHINE

An analysis of various response times for a proposed network multiprocessor database machine is given in this paper. The insights gained should lead to better database machine designs. Our system is based on a principle of "shared-nothing" and aims at supporting large database applications by using parallelism. Our research shows that a highly parallel network multiprocessor DBM should be able to solve the I/O, data processing, and data transfer bottlenecks. A key to overcoming these problems is eliminating the irrelevant data retrieved by the query at each processing stage. For this, we consider both software and hardware techniques, not the use of a multiprocessor architecture alone.