In a multidatabase system it is possible to access and update data residing in multiple databases. The databases may be distributed, heterogeneous, and autonomous. The first part of the thesis provides an overview of different kinds of multidatabase system architectures and discusses their relative merits. In particular, it presents the AMOS multidatabase system architecture which we have designed with the purpose of combining the advantages and minimizing the disadvantages of the different kinds of proposed architectures.

A central problem in multidatabase systems is that of data model heterogeneity: the fact that the participating databases may use different conceptual data models. A common way of dealing with this is to use a canonical data model (CDM). Object-oriented data models, such as the AMOS data model, have all the essential properties which make a data model suitable as the CDM. When a CDM is used, the schemas of the participating databases are mapped to equivalent schemas in the CDM. This means that the data model heterogeneity problem in AMOS is equivalent to the problem of defining an object-oriented view (or object view for short) over each participating database.

We have developed such a view mechanism for relational databases. This is the topic of the second part of the thesis. We discuss the relationship between the relational data model and the AMOS data model and show, in detail, how queries to the object view are processed.

We discuss the key issues when an object view of a relational database is created, namely: how to provide the concept of object identity in the view; how to represent relational database access in query plans; how to handle the fact that the extension of types in the view depends on the state of the relational database; and how to map relational structures to subtype/supertype hierarchies in the view.

A special focus is on query optimization.