Aktive Mechanismen zur Konsistenzsicherung in Föderationen heterogener und autonomer Datenbanken

Heterogeneous information systems are today inevitable for historical and technical reasons. Data is typically maintained on heterogeneous hardware and software platforms and a migration of the legacy systems to a homogeneous environment is not feasible due to high costs and the loss of available functionality. Between existing databases there are global integrity constraints that have to consider both syntactic and semantic data heterogeneity. An active database approach is proposed for the problem of global integrity control. Database federations are modelled as collections of active objects. The active objects are managed by an object-oriented mediator system. The mediator system has to provide a homogeneous interface to heterogeneous components and the functionality to monitor the global consistency among them. This solution allows the external data stores to exist without significant changes but requires additional functionality that has to be provided by the local systems. This addresses a basic principle discussed in this thesis: the conflict between local autonomy and global consistency. The cost of maintaing global constraints is the necessary restriction of the local autonomy. Through specific examples a general schema of trade-offs is derived and it is shown what design decisions are needed for the development of a mediator system and the adaptation of the component systems. It becomes apparent that traditional definitions of consistency in homogeneous database systems are often insufficient, therefore temporal relaxations and partial consistency have to be tolerated in a controlled manner. The global object model of the mediator developed in this thesis is based on a standard data model (ODMG-93/C++) and extensions through which complex dependencies can be described by the use of ECA rules (Event Condition Action). Hence, ECA rules represent our fundamental modeling construct to specify global integrity constraints that are often model-inherent constraints of canonical data models. Due to their expressiveness they are also well suited for weaker consistency criteria, such as temporally relaxed consistency. The submitted thesis describes the implementation of active consistency control mechanisms in a federation of heterogenous relational database systems. The developed mediator system comprises components for the detection, signalling and logging of local database events. These events are interpreted at the global level and processed by a rule management component. The proposed active techniques are applied to the control of autonomous replicated data stores. The specification of complex global integrity constraints is facilitated by an editor tool and can be mapped to ECA rules of the active mediator system. Local schema events may affect the schema consistency and are treated separately. The overall architecture of the mediator system and the interplay of its components are presented.