and NIKOLAOS KOTSIS Capgemini UK Implementation of the data cube is an important and scientifically interesting issue in On-Line Analytical Processing (OLAP) and has been the subject of a plethora of related publications. Naive implementation methods that compute each node separately and store the result are impractical, since they have exponential time and space complexity with respect to the cube dimensionality. To overcome this drawback, a wide range of methods that provide efficient cube

more »

... lementation (with respect to both computation and storage) have been proposed, which make use of relational, multidimensional, or graph-based data structures. Furthermore, there are several other methods that compute and store approximate descriptions of data cubes, sacrificing accuracy for condensation. In this article, we focus on Relational-OLAP (ROLAP), following the majority of the efforts so far. We review existing ROLAP methods that implement the data cube and identify six orthogonal parameters/dimensions that characterize them. We place the existing techniques at the appropriate points within the problem space defined by these parameters and identify several clusters that the techniques form with various interesting properties. A careful study of these properties leads to the identification of particularly effective values for the space parameters and indicates the potential for devising new algorithms with better overall performance.