Optimization of join operations in horizontally partitioned database systems

Publisher Website
Google Scholar
Abstract
This paper analyzes the problem of joining two horizontally partitioned relations in a distributed database system. Two types of semijoin strategies are introduced, local and remote. Local semijoins are performed at the site of the restricted relation (or fragment), and remote semijoins can be performed at an arbitrary site. A mathematical model of a semijoin strategy for the case of remote semijoins is developed, and lower bounding and heuristic procedures are proposed. The results of computational experiments are reported. The experiments include an analysis of the heuristics' performance relative to the lower bounds, sensitivity analysis, and error analysis. These results reveal a good performance of the heuristic procedures, and demonstrate the benefit of using semijoin operations to reduce the size of fragments prior to their transmission. The algorithms for the case of remote semijoins were found to be superior to the algorithms for the case of local semijoins. In addition, we found that the estimation accuracy of the selectivity factors has a significant effect on the incurred communication cost.