CONCLUSIONS

In this chapter we have illustrated the main issues related to the querying of multidimensional data. Querying is strictly connected to the way the data are modeled and particularly to the concepts of data cube, fact table, measure, and dimension. Multidimensional queries have several distinctive features with respect to conventional queries on transactional systems, and in this differentiation aggregation functions and dimension hierarchies play a major role.

We have shown the main proposals to express multidimensional queries, when the data are represented by the relational model (particularly star and snowflake schemes), as well as the approaches based on specifically designed multidimensional data models, usually extensions of the relational one. We have analyzed languages based on an algebra (usually an extension of the relational one), languages based on a calculus (and hence more declarative), and visual languages, usually relying on a collection of algebraic operators. We have shown that two important problems to face, when considering languages based on specific multidimensional models, are the increased complexity of expressing conventional relational operations and the difficulty to symmetrically deal dimensions and measures, i.e., to transform dimensions into measures and vice versa.

We have also focused on the problem of query evaluation, i.e., on how the query expressed in a given language can be translated, using the available materialized views, into an (efficient) evaluation plan which retrieves the necessary information and calculates the required results. This is known in the literature as the query rewriting problem. We have presented some results on aggregate query equivalence and rewriting of aggregate queries that can be used to optimize the evaluation of queries on multidimensional data. Finally, the main indexing techniques for OLAP queries have been illustrated, as well as the contexts of applicability.