Inference techniques for fault tolerant distributed database systems

A data inference approach to increase data availability in distributed database systems is proposed. When the requested data are not accessible owing to network and/or site failures, the database system can infer or approximate them from other accessible database fragments. Two different levels of correlated knowledge are used for inference. In the schema level, correlated knowledge between objects is represented as inference paths. Further, in the instance level, correlated rules are used to represent their detail correlations. In general, inference paths suggest proper objects and directions for data inference. By the selection of proper inference paths, correlated rules can be used to derive the inaccessible information. It is noted that a data inference system can be implemented as a front-end system to an existing distributed database system. It consists of a database fragment availability table which provides the data accessibility information for each site, the inference engine that selects inference paths and rules for inferring unavailable data, and the query modification system which transforms the given query to an alternate one such that all the required database fragments are accessible.<<ETX>>