A Generalisation Approach to Temporal Data Models and their Implementations

Non-temporal data models and their implementations as database management systems (DBMS) capture a single state of the real world, usually the current one. They support modification operations which facilitate the transition from one consistent database state to another. For queries, they assume that the data is synchronous, meaning that all the facts stored in the database are valid at the time instant a query is evaluated. There exist many application domains, however, where it is necessary to reconstruct earlier database states or even store future database states (e. g. for planning, budgets) in parallel. The different database states are stored as temporal data. Such temporal data arises, for example, in financial and insurance applications, in reservation systems and in medical information management. Of course, it is also possible in practice to store timestamps in classical DBMS and model the temporal aspects mentioned above in this way. However, such an approach does not cater for the special semantics of time. Thus, there are many proposals for both relational and object-oriented models as to how the non-temporal data models can be enhanced to support the management of temporal data. Their focus is mainly on extending the data structures and/or the query language. Hardly any of these temporal data models were implemented, even in the form of prototype systems. A more systematic way to define temporal data models is based on generalising a non-temporal data model into a temporal one. Using generalisation means that all constructs of the underlying non-temporal data model — its data structures, operations and integrity constraints — are enhanced to support the management of time-varying data. To show the power of the generalisation approach, this thesis investigates three approaches to managing temporal data, along with the corresponding prototype implementations. The first approach timestamps data by extending the data structures with special timestamp attributes, but, in contrast to existing proposals, uses a generalised query, data definition and data manipulation language. The second approach fully generalises a non-temporal object data model into a temporal one. The resulting temporal object data model TOM does not extend the data structures, but rather uses the notion of temporal object identifiers to timestamp data. In TOM, not only the user data can be timestamped, but also constructs supported by the data model such as collections of objects, types, integrity constraints and so on, since they are also considered to be objects. This temporal data model was implemented as a single-user prototype system. The third approach demonstrates how the extensible nature of object-oriented DBMS can be used directly to support temporal applications through the use of abstract data types. It is shown that while temporal data structures and operations can be accommodated in this way, support for generalised data models and query languages is restricted. These approaches show that a generalised temporal data model is better suited to the modeling and management of temporal data than an extended one, and that generalised data models are implementable. By presenting an evolutionary path from temporal first normal form relations to temporal nested relations, temporal complex objects and temporal object-oriented data models, it is shown that the temporal object data model TOM actually subsumes the extended temporal data models.