Methods for Logical OLAP Design

Abstract

Tiedonanalysointimenetelmien merkitys kasvaa tietomäärien jatkuvasti lisääntyessä. Perinteisesti tietokannoissa tiedon varastoiminen ja hakeminen ovat olleet tärkeimmät toiminnot, mutta nyt painopisteen voidaan nähdä siirtyneen tiedon analysointimenetelmien suuntaan. On-line analytical processing (OLAP) on menetelmä, jossa käyttäjä analysoi tietoa käyttäen moniulotteisia tietokuutioita. OLAP-menetelmää on käytetty lähinnä talousdatan analysointiin, mutta myös muita sovellusalueita on olemassa. Käytettävien moniulotteisten tietokuutioiden suunnittelu on vaikea tehtävä. Esimerkiksi kuution 'harvuudesta' voi aiheutua vakavia ongelmia järjestelmän tehokkuudelle. Tässä työssä on kehitetty funktionaalisten riippuvuuksien käyttöön perustuvia menetelmiä loogiselta rakenteeltaan hyvien tietokuutioiden suunnittelun helpottamiseksi.

As the amount of information is increasing all the time, information modelling and analysis have become essential areas in information management. Storing and retrieving data have earlier been the main functions in databases but the importance of deeper understanding of data has increased during the recent years. The nature of data has also become more complex. Therefore, powerful modelling methods are needed for the data. In addition, the methods have to be user-friendly since the users of data oriented applications are not often database-professionals. The general aim of this work is to develop methods to make database design and management easier and more efficient for all users, both database professionals and people with no prior experience with databases.

We have studied methods to support logical design of OLAP (On-Line-Analytical Processing) cubes. The methods give a good basis for implementing software tools that partly automate the logical design process. These methods are needed since it is commonly noticed that logical design of OLAP cubes is a complex process that requires good knowledge on both application area and databases. Good logical structure of an OLAP cube is important, because a bad design can lead, for example, to an extremely sparse cube and to such a need for storage space that is not possible to achieve in practice. As a solution, we give a method for estimating the structural sparsity of OLAP cubes and a normal form to reduce sparsity risks. Moreover, synthesis and decomposition algorithms for producing normalised OLAP cubes are developed. Hierarchical dimensions, which enable the user to analyse data on different levels of aggregation, are essential for OLAP. Hierarchies can arise from the attribute hierarchy (e.g. day, month, year) or from the relationship between the instances of two attributes (e.g. employee, manager). We study what kinds of hierarchy structures are desirable with respect to correct aggregations and efficient calculations. To represent logical OLAP schemata, a dependency-based modelling method has been developed. This method enables the user to describe concepts and their relationship to each other explicitly. The OLAP cube should be complete and minimal with respect to the user's queries. To define what data should be taken into account when constructing an OLAP cube, we give two methods based on query information. One applies the intensional concept theory and a query method based on it. The other uses MDX queries that the user poses against a base cube representing the contents of the data warehouse. If real queries are available, they can be used as input.