View materialization issues in object-oriented databases.

Abstract

Recent advances in information technology introduced a need for techniques to integrate heterogeneous sources, to cache and re-use results of queries across multiple information sources (data warehousing), to support customized interfaces to shared data, and to integrate such mechanisms with the powerful constructs of the object-oriented programming model. Object-oriented database (OODB) view technology can help provide such techniques. A view is a query that is stored and given a name by which it can be used in other queries. Typically a view's contents are derived using the view's stored query. View materialization, i.e., the generation of derived extents of views, improves the performance of queries by eliminating the need for such recomputation. However, it requires the maintenance of the contents of materialized views in the face of updates. This thesis proposes a methodology for the support and maintenance of materialized OODB views. We have implemented a prototype of this methodology in the context of the MultiView OODB view system, one of the first view management systems for object-oriented databases and (to the best of our knowledge) the first to support updatable incrementally materialized object-oriented views. The primary contribution of this work is the development of algorithms that exploit OO characteristics for the incremental maintenance of materialized OODB views. In particular, we address two potential inefficiencies to which update propagation algorithms are subject--the propagation of updates to irrelevant derived classes and the propagation of self-cancelling updates. We present cost models and experimental results that demonstrate that our techniques of hierarchical registration of the dependency of virtual classes on properties invoked in their predicates and derivation-ordered propagation successfully address these problems. We also propose a new Satisfiability Indicating Multi-Index (SMX) organization specifically tailored to optimize the maintenance of views defined by selection queries along aggregation paths (path query views). A key characteristic of our SMX solution is to maintain minimal partial information indicating whether or not the endpoints reachable from an object satisfies the query predicate. Our cost models confirm that the SMX dramatically improves upon the performance of traditional index structures with respect to the problem of path query view maintenance.