Physical Data Modeling for Multidimensional Access Methods

Despite the fact that the database community has proposed a vast number of indexing methods over the years, no standard physical data model has been established like it has been achieved on the conceptual and logical level. How to optimize a given data model by using various indexing methods is still the ‚trade secret‘ of the database administrators. Only recently, some approaches have been tried to make this knowledge available to the normal database user by easy to use optimization tools (e.g., AutoAdmin-Tool of MS SQL Server 7.0). In addition, physical data modeling has concentrated on one-dimensional access methods, since these were the only ones available in commercial database management systems. As multidimensional access methods (MDAMs) are making their way from the research labs into commercial products, a general physical data model should also take MDAMs into account, especially since MDAMs have a high potential to improve processing in important application domains like OLAP, Data Mining, or Archiving Systems. Our research in this field concentrates on providing rules and heuristics for optimal physical data modeling with multidimensional access methods. Currently we are focusing on the application domain of relational OLAP (ROLAP). MDAMs have a high potential in ROLAP since most queries result in multiattribute restrictions on a table [MZB99]. Arguing that physical data modeling is superfluous in the presence of MDAMs because one could just index all important attributes with one MDAM neglects the fact that the practical limits on the dimensionality of MDAMs lies around ten dimensions. As consequence, physical data modeling for MDAMs does not address the question of which index type to use in the first place, but the question of which attributes to select for indexing.