Modeling and retrieval techniques for generalized bitemporal databases

Temporal databases store information as it changes over time. In current research, two time dimensions are given particular importance: the valid time, when a fact is true in the modeled reality, and the transaction time, when a fact is known to the database system. Bitemporal databases (BTDBs) include both time dimensions. The contributions of this thesis involve identifying several problems with existing BTDBs, proposing a model that overcomes these problems, and specifying indexing techniques that allow efficient retrieval using this model. We show that it is not sufficient to just combine valid time and transaction time representation methods to represent a BTDB, because several problems can occur: the lost information problem, the ambiguity problem, and the priority specification problem. In this thesis, we propose a generalized bitemporal database (GBTDB) model to overcome those problems. The GBTDB model is a pure append-only model, in which no physical changes are allowed. The effect of deletions and updates are realized by appending information that invalidates previous values. In the GBTDB model, a computational procedure is needed to compute the currently valid information, which we call All-Valid-Versions (AVVs). In order to efficiently process all types of bitemporal queries in the GBTDB, efficient algorithms to calculate and retrieve AVVs are critical. Because BTDBs are often very large, index structures are needed for the efficient processing of bitemporal database queries. Such an index structure should be able to index various combinations of key, valid time, and transaction time. In this thesis, we first specify and discuss the problems that may occur in BTDBs and then formally define the GBTDB model and the concept of AVVs. Then we provide algorithms to retrieve it AVVs incrementally in both forward and backward directions. An integration bitemporal index structure that combines the indexing of key, valid time, and transaction time, called IBI-Tree, and its variation, called $IBI\sp+$-Tree are proposed. The $IBI\sp+$-Tree reduces the space requirement dramatically compared to the IBI-Tree. Performance evaluation of the index structures is then presented.