Browse Theses

This thesis presents the design and analysis of a collection of algorithms to support triggers, inference rules, and derived data objects (e.g. views) in relational database systems. A view maintenance algorithm, is a method for maintaining and incrementally updating a physically stored copy of a database view. A new view maintenance algorithm called Rete view maintenance (RVM) is proposed in this thesis. RVM is based on the Rete Network, a type of discrimination network used to test rule conditions in forward-chaining rule interpreters. A collection of algorithms is also proposed to allow maintenance of materialized aggregates and aggregate functions.

By keeping a stored copy of a view up-to-date using a view maintenance algorithm, it is possible to process view queries directly using the copy. The conventional way to process queries against views is to use query modification, whereby a view query is translated into an equivalent query that refers only to the base relations. A performance analysis is presented which compares the average cost of a view query for these two alternatives for different view types.

A related performance analysis is also presented comparing the costs of different algorithms for querying database procedures. The database procedures analyzed are made up of one or more database queries stored in the field of a record. The value of a database procedure is the result of executing the query or queries in its definition. Three different algorithms for processing queries against database procedures are evaluated. As in the case for views, the average query cost for each algorithm is compared.

Finally, enhancements to the rule sublanguage of the POSTGRES database management system are proposed to increase the power of the language and to simplify implementation of rule-based applications. Methods are presented for implementing the new language features efficiently using view maintenance methods combined with techniques for indexing rule predicates.