Browse Theses

In the last few years, there has been significant interest in the design of incremental methods to improve the performance of view maintenance. Despite that, very little analysis or experimentation supports the predominant view that incremental methods are more efficient than their non-incremental counterparts. We argue that the performance of incremental view maintenance depends on system aspects of the database, such as the availability of indices, the sizes of the relations involved, and the sizes of the database updates. We also argue that the database query optimizer is a reasonable component of the database system to decide, at the time of view maintenance, whether a view is to be maintained incrementally or not, because the query optimizer has knowledge of, and access to, all of the parameters that may affect this choice.To support this argument, we have built the RHODES database query optimizer that supports change propagation and view maintenance for relational queries. In addition to traditional optimizations, RHODES is also responsible for the generation of the queries to be executed in order to support view maintenance. As there may be many different ways to maintain a view incrementally, the choice of which one to use may affect the performance of incremental view maintenance. Moreover, different maintenance queries are amenable to different optimizations. In this thesis, we present a repertoire of maintenance-specific optimizations, especially in the presence of key constraints and foreign key references. The underlying data model we use is relational algebra with multiset semantics. Experimental validation of the above claims has been conducted using the TPC-D benchmark database on the DB2 Parallel Edition.