OLAP applications use precomputation of aggregate data to improve query response time. While this problem has been well-studied in the recent database literature, to our knowledge all previous work has focussed on the special case in which all aggregates are computed from a single cube (in a star schema, this corresponds to there being a single fact table). This is unfortunate, because many real world applications require aggregates over multiple fact tables. In this paepr, we attempt to fill this lack of discussion about the issues arising in multi-cube data models by analyzing these issues. Then we examine performance issues by studying the precomputation problem for multi-cube systems. We show that this rpoblem is significantly more complex than the single cube precomputation problem, and that algorithms and cost models developed for single cube precomputation must be extended to deal well with the multi-cube case. Our results from a prototype implementation show that for multi-cube workloads substantial performance improvements can be realized by using the multi-cube algorithms.
展开▼
