SIGMAFDB: Overview of the Magdeburg-Approach to Database Federations

Our research activities in the eld of federated database systems (FDBS) are closely related to the SIGMAFDB-project which is founded by the German state Sachsen-Anhalt since 1995. In this extended abstract, we present our view on federated database systems and sketch some of the main results of the work of our group. We brie y discuss di erent research aspects and implementation activities. Due of the speci c topic of this workshop, we assume the familiarity with the basic notions of federated databases. Therefore, we omit a general motivation for federated database systems. In the following, we present our application scenarios very brie y. In the rst two years of the project, we focused on integration scenarios in factory planning. In this eld, di erent speci c application programs are used for optimizing machine con guration and transport facilities in factories. The applications are developed independently. Although some kind of computerbased interoperability is necessary, nowadays interoperability is performed manually. In most cases, interoperation is uni-directional and primarily based on le-exchanges. In consequence, data inconsistencies may occur which means additional work for the factory designer. The result of analyzing the systems is a formal description of the used le formats in SGML and an integrated schema. Beside this, we developed a \hard-coded" demonstration prototype for illustrating problems and features of database integration using federated database systems. Our second integration scenario was motivated by common work with our partners in the bio-computer science research group of our institute. In this eld, we found again highly heterogeneous databases which contain bio-molecular data that represent results of thirty years of experimental research. The integration of these partly overlapping databases promises a new quality of databases and enables new applications to access heterogeneous data in a uniform and transparent way. We have to explicitly mention two speci c aspects of this scenario. Firstly, most databases are based on les and often provide a WWW interface. Secondly, only reading access is possible. In this context, local (data) autonomy is important and cannot be relaxed. A more detailed discussion about the application of FDBS to this area and a description of the implementation of the prototype BIO-Bench can be found in [9]. Our third (and newest) integration scenario is motivated by the Global-Info-project which aims at supporting digital libraries by federation services. As in the scenarios before, di erent information systems, mainly with WWW interfaces, are available. These systems provide related and overlapping informations. Nowadays, a user has to access some or many information systems for information retrieval using his personal knowledge (meta-data) to nd the correct (and needed) information. For example, even in the computer science society, it is not trivial to assign the right real-world person to the string \M. Scholl". Another problem is duplicate representation which is very hard in manual work. For that, a federation layer can support the user in his work with services and meta-data, e.g. author-publicationseries-networks. Our common view to federated database systems is a tightly coupled approach based on the ve-level schema architecture of Sheth and Larson [13]. This is motivated by the requirement of data consistency which can only be guaranteed in a tightly coupled environment. A federated database should correctly re ect the modeled real world. Inconsistencies between the databases concerning to same real world entities have to be removed. Of course, this is contradictory to the requirement of local autonomy, but