Adding a Performance-Oriented Perspective to Data Warehouse Design

Data warehouse design is clearly dominated by the business perspective. Quite often, data warehouse administrators are lead to data models with little room for performance improvement. However, the increasing demands for interactive response time from the users make query performance one of the central problems of data warehousing today. In this paper we defend that data warehouse design must take into account both the business and the performance perspective from the beginning, and we propose the extension to typical design methodologies to include performance concerns in the early design steps. Specific analysis to predicted data warehouse usage profile and meta-data analysis are proposed as new inputs for improving the transition from logical to physical schema. The proposed approach is illustrated and discussed using the TPC-H performance benchmark and it is shown that significant performance improvement can be achieved without jeopardizing the business view required for data warehouse models.