Knowledge-based retrieval on a relational database machine

The central focus of the doctoral research which is described in this dissertation has been the efficient retrieval of records from very large databases in applications where the criteria for description-matching require deductive inference over a domain-specific "knowledge base". The approach on which this research is based, while offering considerable power and generality, requires the execution of operations which are quite expensive on a conventional computer system. The solution which has been pursued has involved the design of a specialized non-von Neumann machine, portions of which would be implemented in VLSI, which permits the highly efficient evaluation of certain operators of a relational algebra of particular importance to the computational task of logical satisfaction. The architecture permits an O(log n) improvement in time complexity over the best known evaluation methods for these operators on a conventional computer system, and appears to offer significant potential advantages in many practical database management applications. A working system which matches KRL-like descriptions {Bobrow and Winograd, 1977}, using a very simple knowledge base, has been implemented, with the database machine emulated in software at the macro-instruction level. The essential mechanisms of this system are described in the dissertation, along with the architecture of and hardware algorithms employed in the relational machine. An analysis of the time complexity of these algorithms is also presented.