Physical database design in multiprocessor database systems
Shahram Ghandeharizadeh, David J. DeWitt · Minds at UW (University of Wisconsin) · 1990
In shared-nothing multiprocessor database machines, relations are horizontally declustered across multiple processors in order to obtain a lower response time and a higher throughput from the system. Several alternative strategies exist for horizontally declustering a relation. However, the performance tradeoffs of these declustering strategies with respect to the different storage and access structures have not been previously examined. Thus, a database administrator for such a system has no choice but to guess how many processors a relation should be declustered across and what declustering strategy is most appropriate. These are important decisions since they have a significant impact on the response time and throughput of the system. In this thesis, we evaluate the performance of the alternative declustering strategies for different selection query types in a multiuser environment and quantify the tradeoffs associated with each organization in the context of the Gamma database machine. Based on these results, we propose a new declustering strategy termed the multidimensional declustering strategy that is a better alternative than the existing declustering strategies. We present the design of this declustering strategy and compare its performance to the current declustering strategies provided by Gamma.