ARCube: Supporting Ranking Aggregate Queries in Partially Materialized Data Cubes

Supporting ranking queries in database systems has been a popular research topic recently. However, there is a lack of study on supporting ranking queries in data warehouses where ranking is on multidimensional aggregates instead of on measures of base facts. To address this problem, we pro- pose a query execution model to answer di®erent types of ranking aggregate queries based on a uni¯ed, partial cube structure, ARCube. The query execution model follows a candidate generation and veri¯cation framework, where the most promising candidate cells are generated using a set of high-level guiding cells. We also identify a bounding princi- ple for e®ective pruning: once a guiding cell is pruned, all of its children candidate cells can be pruned. We further address the problem of e±cient online candidate aggrega- tion and veri¯cation by developing a chunk-based execution model to verify a bulk of candidates within a bounded mem- ory bu®er. Our extensive performance study shows that the new framework not only leads to an order of magnitude performance improvements over the state-of-the-art method, but also is much more °exible in terms of the types of rank- ing aggregate queries supported.