On effective data placement based on temporal properties of data

Abstract

An object in the real world has a life cycle of creation, evolution, and extinction. On the life cycle, the content which describes the object changes constantly and the time of the changes as well as the content itself is a crucial part of the information. Thus, time is an essential property of almost every real world entities. The temporal characteristics of data objects can be managed effectively by using temporal databases. Temporal databases allow users to record and retrieve time-varying data objects. Since temporal databases usually manage a huge amount of underlying data objects, efficient disk accesses are essential for fast response time in temporal query processing. Data clustering is one of the most effective techniques that can improve performance of temporal database systems. A clustering method of a database system requires a proper clustering criterion and an effective measure for the criterion which is a basis of the decision on the selection of data objects to be clustered together. However, clustering measures for conventional data objects are not appropriate to temporal data objects because it is important to exploit temporal properties of underlying data objects and temporal queries as data clustering criteria. In this thesis, we propose a data clustering measure, called temporal affinity, that can be used for effective temporal data clustering. The temporal affinity, which is based on the semantics of temporal operators, reflects the closeness among temporal data objects with respect to temporal query processing. We perform experiments to show the effectiveness of the proposed temporal data clustering measure. The experimental results indicate that a data clustering method based on the temporal affinity works better than other methods.