Interoperability in Data Warehouses

Abstract

SYNONYMS Data warehouse integration DEFINITION The term refers to the ability of combining the content of two or more heterogeneous data warehouses, for the purpose of cross-analysis. This need emerges in a variety of practical situations. For instance, when different designers of a large company develop their data marts independently, or when different organizations involved in the same project need to integrate their data warehouses. Data Warehouse interoperability is a special case of the general problem of database integration, but it can be tackled in a more systematic way because data warehouses are structured in a rather uniform way, along the widely accepted concepts of dimension and fact. As it happens in the general case, different degrees of interoperability can be pursued by adopting standards and/or by applying reconciliation techniques, likely specific for this context. The problem is becoming increasingly relevant with the spreading of federated architectures. Nevertheless, it has been the focus of a few systematic works and numerous open problems remain to be solved. HISTORICAL BACKGROUND In spite of its relevance, the problem of data warehouse integration has received little attention so far. Conversely, the general problem of databases integration has been studied in the literature extensively and several aspects, both at scheme and instance level, have been deeply investigated, such as the automatic matching of terms and the resolution of structural conflicts (see [8, 12] for surveys on these topics). In the specific context of data warehouses, Kimball [7] has identified the problem for the first time: he has investigated the integration of heterogeneous dimensions in a scenario of data warehouse design and has introduced the informal notions of dimension conformity. Intuitively, two dimensions are conformed if their share some information in a consistent way. This is an important requirement in drill-across queries, which are basically joins of different facts over common dimensions. The notion of conformity has been formalized and extended by Cabibbo and Torlone in the context of data mart integration [3] under the name of dimension compatibility: they have demonstrated that this property gives the ability to perform correct drill-across queries over heterogeneous data marts. An issue related to the integration of data warehouses, which has been studied in the context of statistical databases, is the derivability of summary data. This notion has been defined by Sato [14] as the problem of deciding whether a summary data (which is, in a statistical database, the counterpart of a fact table) can be inferred from another summary data aggregated in a different way. The concept has been extended by Malvestuto [9], by considering the case in which the source is composed by several heterogeneous data sets: he proposes an algebraic approach to this problem and provides some necessary and sufficient conditions of derivability. Unfortunately, statistical databases have some similarity with multidimensional databases, but also some important diversities: this makes the application of these approaches to data warehouses not easy. Some related work has been done on the problem of integrating a data warehouse with external data stored in XML [6] and in object-oriented [11] format, but just a few works have been devoted to the specific problem of the interoperability between heterogeneous data warehouses. They will be discussed in the following section. While current commercial tools do not provide a complete support for data warehouse interoperability, they offer facilities that can be very useful in this framework, such as metadata import/export (using XML) and