Methodical construction of database management systems

Although more and more database management systems are being developed, constructing such systems in an engineering fashion remains an open problem. An engineering attitude towards construction not only tells how to implement a database management system, but also tries to minimize construction time. Hence, the prime objective of an engineering-style construction approach is construction efficiency. This thesis presents a methodology for constructing centralized database management systems. The construction method proposes two major building blocks in order to support the construction of database management systems: an architecture model and a model of the construction process. The architecture model is a prerequisite for the development and representation of the structure of a database management system. The construction process structures the development of such a system. Reuse techniques are applied in all phases of the construction in order to achieve greater construction efficiency and to consider results of previous constructions. The thesis proposes a client/server-like model for representing architectures (the broker/services model). Services are the functionality that is implemented by parts of the database management system. Brokers are active agents that are responsible for providing services and that request services from each other. Furthermore, a construction process is proposed comprising phases for analysis, specification, design, and implementation. Architectures of previous constructions as well as specifications are subject to reuse in early phases of this process. Apart from architectures, further kinds of reusable artifacts are identified for each phase. Most importantly, the thesis introduces constructs for representing and classifying reusable artifacts in domains for which large varieties of realizations exist. Dimensions are attached to each such domain; they identify design decisions to be made. Each dimension defines a set of strategies, where each strategy reflects an alternative way to realize a specific functionality. Strategies are formally specified. Knowledge about how to implement strategies is given in the terms of techniques. Techniques are formally specified complex classes in the sense of object-oriented programming; they are reused during the implementation phase. For each phase and its specific reusable artifacts it is shown how finding and selecting candidates for reuse can be supported. This work investigates the construction of one specific aspect of database managment systems: transaction management. The thesis describes a domain analysis of transaction management. The major result of this analysis is a classification of concurrency control approaches in terms of dimensions and strategies. The thesis also presents a specification language for transaction models. It then shows how transaction management subsystems are constructed in the context of the broker/services model, the specification language, and the classification in terms of strategies. Finally, the design of a support system for the construction of database management systems is illustrated, and a sample construction is outlined.