| Distributed database systems can yield significant cost and response time advantages over centralized systems for geographically distributed organizations. However, inappropriate design can result in high cost and poor response time. The design of efficient distributed databases is a complex task involving two interdependent problems: data allocation (where to place the data) and operation allocation (how and where to process the data). To be effective, a distributed database design approach must comprehensively treat data and operation allocation as a unified whole. Furthermore, trade-offs among retrieval and update requests must be evaluated since replicating data at multiple nodes to enhance retrieval efficiency increases update costs as consistency among multiple copies of data must be maintained. In this research, we develop comprehensive distributed database design models that include replicated data, retrieval and update queries, and efficient operation allocation strategies (e.g., join node selection, join order, reduction by semijoin). Due to the combinatoric nature of the problem, conventional solution algorithms are intractable for reasonably sized problems. Hence, we develop genetic algorithms to solve them. We experimentally compare our models with existing approaches. We conclude that data replication, reduction by semijoin, join node selection, join order are significant factors in the overall performance of a distributed database design. Finally, we validate our performance models using simulation. |
