Managing schema change in an heterogeneous environment.

摘要

Change is inevitable even for persistent information. Effectively managing change of persistent information, which includes the specification, execution and the maintenance of any derived information, is critical and must be addressed by all database systems. Today, for every data model there exists a well-defined set of change primitives that can alter both the structure (the schema) and the data. Several proposals also exist for incrementally propagating a primitive change to any derived information (or view). However, existing support is lacking in two ways. First, change primitives as presented in literature are very limiting in terms of their capabilities allowing users to simply add or remove schema elements. More complex types of changes such the merging or splitting of schema elements are not supported in a principled manner. Second, algorithms for maintaining derived information often do not account for the potential heterogeneity between the source and the target. The goal of this dissertation is to provide solutions that address these two key issues.; The first part of this dissertation addresses the challenge of expressing a rich complex set of changes. We propose the SERF (Schema Evolution through an Extensible, Re-usable and Flexible) framework that allows users to perform a wide range of complex user-defined schema transformations. Our approach combines existing schema evolution primitives using OQL (object query language) as the glue logic. Within the context of this work, we look at the different domains in which SERF can be applied, including web site management. To further enrich our framework, we also investigate the optimization and verification of SERF transformations.; The second part of this dissertation addresses the problem of maintaining views in the face of source changes when the source and the view are not in the same data model. With today's increasing heterogeneity in information structure, it is critical that maintenance of views addresses the data model boundaries. However, view definitions that go across data models are limited to hard-coded algorithms, thereby making it difficult to develop general maintenance algorithms. We provide a two-step solution for this problem. We have developed a cross algebra, that defines views such that there is no restriction that forces the view and the source data models to be the same. We then define update propagation algorithms that can propagate changes from source to target irrespective of the exact translation and the data models. We validate our ideas by applying them to translation and change propagation between the XML and relational data models.