Development of a fault tolerant distributed database via inference

The authors report on the experience of building a knowledge-based distributed database testbed on top of a commercial relational database to experiment with semantics for fault tolerance. This work sets out to test the philosophy that not only can syntactic redundancy (replication) be exploited to improve fault tolerance, but that most data are correlated, containing redundant information at the semantic level as well. The database must be engineered to make use of the inference engine to infer the inaccessible from the accessible data. The authors provide an overview of the architecture and discuss the data inference system, the knowledge schema, caching, the commercial database server, and database error handling.<>