Unifying temporal data models via a conceptual model

To add time support to the relational model, both first normal form (1NF) and non-1NF data models have been proposed. Each has associated advantages and disadvantages. For example, remaining within 1NF when time support is added may introduce data redundancy. On the other hand, well-established storage organization and query evaluation techniques require atomic attribute values, and are thus intended for 1NF models; utilizing a non-1NF model may degrade performance. This paper describes a new temporal data model designed with the single purpose of capturing the time-dependent semantics of data. Here, tuples of bitemporal relations are stamped with sets of two-dimensional chronons in transaction-time/valid-time space. We use the notion of snapshot equivalence to map temporal relation instances and temporal operators of one existing model to equivalent instances and operators of another. We examine five previously proposed schemes for representing bitemporal data: two are tuple-timestamped 1NF representations, one is a Backlog relation composed of 1NF timestamped change requests, and two are non-1NF attribute value-timestamped representations. The mappings between these models are possible using mappings to and from the new conceptual model. The framework of well-behaved mappings between models, with the new conceptual model at the center, illustrates how it is possible to use different models for display and storage purposes in a temporal database system. Some models provide rich structure and are useful for display of temporal data, while other models provide regular structure useful for storing temporal data. The equivalence mappings effectively move the distinction between the investigated data models from a semantic basis to a display-related or a physical, performance-relevant basis, thereby allowing the exploitation of different data models by using each for the task(s) for which they are best suited.