(A) performance evaluation of the parallel joun algorithms on a distributed-memory multiprocessor database computer

Abstract

During the past decades, the effective and efficient management of large quantities of data has been a common problem found in most large business, scientific, engineering, and military applications. There has been a considerable amount of research and development effort in the design and development of multiprocessor database computers that are tailored for processing large databases in parallel. Also a large amount of effort in the design and performance study of parallel algorithms on the multiprocessor has been done for efficiently performing complex database management functions, such as the join operation. In this thesis, we present the performance evaluation of parallel join algorithms running on the multiprocessor database computer. First, we design and implement the multiprocessor database computer as a type of he distributed-memory multiprocessor. The overall performance of the multiprocessor database computer is measured and evaluated using well-known benchmarking method, and then good results are shown compared with other systems. Next, a various parallel algorithms to carry out the join operation on the multiprocessor database computer is presented and developed by the structural way, and the performance of each of the algorithms with different tuple distribution and local joining policies, addition of bit vector filters, and non-uniformly distributed join attribute values are studied. the performance evaluations are also conducted by an experimental way on our multiprocessor database computer, and the results are compared with the analytic results. As a result of the performance evaluation, we show that thee is not just one overall best performing join algorithm, but different algorithms score the best performance, depending on the characteristics of the data participating in the join operation.