Performance of Grace Hash Join Algorithm on the Ksr-1 Multiprocessor: Evaluation and Analysis Performance of Grace Hash Join Algorithm on the Ksr-1 Multiprocessor: Evaluation and Analysis

In relational database systems, the join is one of the most expensive but fundamental query operations. Among various join methods, the hash-based join algorithms show great potential as they lend themselves for parallelization. Although performance of the hash join algorithm has been evaluated for many architectures, to the best of our knowledge, it has not been evaluated for the COMA memory architecture. This paper evaluates and analyzes the performance of hash join on the KSR-1 shared-everything multiprocessor system which has the COMA memory structure. The following performance related issues are presented in this paper for an implementation of the Grace hash join algorithm: double buuering, data partition and distribution, potential parallelism, and synchronization. The performance evaluation results for the above are presented and analyzed. The results and analysis indicate that the ALLCACHE memory organization of KSR-1 is beneecial for parallelizing the Grace hash join algorithm.