Learning techniques for query optimization in federated database systems

The architecture of ADZE, an adaptive query optimizing system for federated databases, is presented. ADZE applies an explanation-based learning technique to learn from failures. It creates selection rules that guide the optimizer in situations where it has previously failed in selecting an optimal strategy. Other learning techniques, such as empirical learning and caching of values, are also used by the system in the process of creating and refining selection rules. The relevance of explanation-based learning to this type of application is discussed