An Analysis of Additivity in OLAP Systems

Abstract

Accurate summary data is of paramount concern in data warehouse systems; however, there have been few attempts to completely characterize the ability to summarize measures. The sum operator is the typical aggregate operator for summarizing the large amount of data in these systems. We look to uncover and characterize potentially inaccurate summaries resulting from aggregating measures using the sum operator. We discuss the effect of classification hierarchies, and non-, semi-, and fully-additive measures on summary data, and develop a taxonomy of the additive nature of measures. Additionally, averaging and rounding rules can add complexity to seemingly simple aggregations. To deal with these problems, we describe the importance of storing metadata that can be used to restrict potentially inaccurate aggregate queries. These summary constraints could be integrated into data warehouses, just as integrity constraints and are integrated into OLTP systems. We conclude by suggesting methods for identifying and dealing with non-and semi-additive attributes.