Rule processing issues in expert database system

Database Management Systems (DBMSs) are being extended with rule management facilities to support rule-based, data-intensive decision making, resulting in Expert Database Systems (EDSs). In this research, several rule processing techniques and strategies were developed that have important impact on the critical issues in the design and implementation of EDSs. The contributions of this thesis are as follows: (1) A rule processing framework was proposed for non-chained rules based on two new data constructs, the condition pattern relation and the join pattern relation. The condition pattern relation is a partial materialization technique that maintains rule-relevant data elements so that activation of the rule can be done without reading the base data. This construct is very efficient when the base relation is large or not properly indexed. The join pattern relation is a join indexing technique that enables the elimination of expensive "false drops"--repeated accesses of data elements that do not trigger the rule. Several rule processing strategies based on the rule processing framework were modeled. (2) A selective materialization paradigm was introduced for chained rules to optimize rule processing costs. A decomposition-based algorithm was developed for materialization decisions so that the cost of rule processing is minimized while satisfying constraints on query response time. Theoretical models and experiments were conducted to illustrate the cost savings potential. (3) A prototype rule system was implemented on top of the Postgres DBMS to demonstrate the validity and practically of the selective materialization paradigm. The prototype system is composed of three main components: the rule transformation module that converts logical rules into physical rules that incorporate materialization decisions; the update propagation module that maintains materialized data incrementally when base data are updated; and the query propagation module that recomputes non-materialized data when they are requested by a retrieval query. Empirical results showed that rule processing costs under selective materialization can be significantly lower than both the "materialize none" and "materialize all" approaches.