Efficient Implementation of Data Cubes via Materialized Views

Abstract

Data cubes are specialized database management systems designed to support multidimensional data for such purposes as decision support and data mining. For a given mix of queries, we can optimize the implementation of a data cube by materializing some projections of the cube. A greedy approach turns out to be very effective; it is both pas a function of the number of possible views to materialize and guaranteed to come close to the optimum choice of views. The work reported here is a summary of results appearing in the following two papers: V. Harinarayan, A. Rajaraman, and J. D. Ullman, "Implementing data cubes effciently." To appear in 1996 SIGMOD. An extended version is available by anonymous ftp from db.stanford.edu as pub/harinarayan/1995/cube.ps. H. Gupta, V. Harinarayan, A. Rajaraman, and J. D. Ullman, "Index selection for OLAP." Available by anonymous ftp from db.stanford.edu as pub/hgupta/1996/CubeIndex.ps.