A Comparison of Distributed Database Design Models

Abstract

Although numerous database design models and solution algorithms have been developed, there has been little work that compares and evaluates these models. Lack of such work has left us with several questions: Do the more comprehensive models actually result in better solutions than the simpler models? If so, what makes them better? Are they better under all conditions or only under certain conditions? Are there trade-offs between data redundancy and sophisticated operation allocation strategies? In this paper, we systematically compare and evaluate several distributed database design models in terms of total operating cost and average response time under various conditions. We vary the relative frequency of update queries and selectivities of queries. The results demonstrate that replication, join node selection, join order, and reduction by semijoin, all have significant impact on the efficiency of a distributed database system. Replication was most effective for retrieval intensive and high selectivity situations. Join node selection, join order, and reduction by semijoin were most effective for balanced retrieval/update and low selectivity situations. The results also suggest that there are trade-offs between total operating cost and average response time design criteria.