One of the most promising technologies being used for constructing knowledge-base systems is the integration of programming languages based on the logic paradigm, Prolog, for example and relational DBMSs. In theory, at least, it provides a base on which development systems for constructing very large knowledge-base systems could be implemented. However, to achieve this goal in practical terms, improvements in performance by several orders of magnitude over existing systems must be obtained. Toward this performance improvement, an examination is made of the compilation of logic programs and their subsequent storage, retrieval, and execution in the specified context. An inspection is made of the relationship between relational engines as found in relational DBMSs and the Warren Abstract Machine (WAM), the most efficient and powerful model used for the construction of Prolog compilers. The development of the Educe system is used to illustrate implementation issues.>

Educe is a logic programming system for handling large knowledge bases. It was constructed by fully integrating the logic programming language Prolog and the relational data base management system Ingres. Educe uses a hybrid strategy for the evaluation of queries. This strategy is based on two contrasting strategies. The strategy known as sets retrieval, transforms recursive and non-recursive queries into a form suitable for evaluation by a relational data base management system. The other strategy, known as one-tuple-at-a time, evaluates queries by imitating the evaluation strategy of the programming language Prolog. In earlier versions of Educe, users selected the strategy by using two different query languages. In order to remove this responsibility from the user, algorithms to map expressions from either of the languages into the other were implemented and added to Educe. This paper briefly reviews the implementation of both evaluators and the mappings compares the basic strategies of evaluation, and then proceeds to explain Educe's own strategy.