View-based solutions for schema evolution problems: Application interoperability and view synchronization.

Abstract

Research carried out for this dissertation examines aspects of two important problems, namely the application interoperation problem and the view evolution problem, both caused by schema changes of the underlying information sources (ISs). The application interoperation problem occurs when a schema change demanded by one user has an adverse effect on other existing applications using the same IS. The view evolution problem occurs when an IS changes its schema and renders the views derived from it undefined. Solution approaches to them by utilizing and extending view technologies are provided. Specifically, this dissertation makes several major contributions. Object-oriented views allow interoperating applications to evolve while minimizing or even eliminating the impact of the schema changes on a shared schema on other existing applications. When a system is plagued with redundant and obsolete views and versions, its performance degrades. One major contribution of the dissertation are the design and implementation of the Schema Version Removal tool (SVR) that removes excessive build-up views and versions for the optimization of such systems. Another contribution is the provision of a practical Transparent Object Schema Evolution system (TOSE) for the application interoperability problem. TOSE is based solely on standard and hence available view and in-place schema change technologies, whereas previous solutions required advanced non-standard OO view capabilities. Advanced applications such as web-based information services create and maintain tailored information repositories (views) by gathering data from a large number of internetworked dynamic ISs. Evolvable views are a must in such dynamic environments. We propose the first solution, called evolvable view environment (EVE) framework, to the problem of maintaining views when the underlying ISs change their schemas. One unique contribution of EVE is the extension of the conventional view specification language SQL to express relaxed query semantics, called E-SQL, and thus allow us to direct the evolution process. Another contribution are the design and implementation of view synchronization algorithms that evolve views under IS capability changes, while assuring that they continue to meet the desired E-SQL semantics. The third contribution is the design of the QC model for selecting the best query rewriting among a possibly large set of rewritings.