Design and Analysis of Index Structures in MultiVersion Data Warehouses
Khaled Jouini, Geneviève Jomier · Base Institutionnelle de Recherche de l'université Paris-Dauphine (BIRD) (University Paris-Dauphine) · 2009
A MultiVersion Data Warehouse (MVDW) is a Data Warehouse (DW) using versioning to cope with the slowly evolving nature of analysis dimensions as well as to support what-if analyses. In such a DW, several DW versions coexist. Each DW version is composed of data extracted from data sources during a certain time period, or of computed hypothetical data, generated to analyze the outcomes of an hypothetical strategic decision. In a MVDW, data may evolve along different version branches and queries may involve multiple DW versions. The efficiency of query processing is crucial in a DW environment and requires that DW designers be aware of tradeoffs among index structures. This paper defines a framework for understanding, designing, analyzing and comparing index structures for MVDW. First, the paper extends index structures designed for data with linear evolution to handle data with branched evolution. Next, the paper defines an analytical method for understanding the behavior of the different index structures. The analysis allows determining the most suitable index structure, for given data and application characteristics.