Eecient Processing of Queries Containing User-deened Predicates

Query optimizers often have diiculties with the treatment of user-deened functions. In particular, the optimization of joins involving complex, user-deened predicates rather than just simple arithmetic comparisons may lead to query plans of poor quality. In this paper, we identify a class of user-deened functions that can be included in queries in such a way that eecient query optimization is still possible. For this class of functions, a join between two sets R and S, for example, could still be computed in time signiicantly less than O(jRj jS j). To achieve this goal we introduce the concept of the-function, an operator to process each set separately with respect to the user-deened func-tion(s) being used. These-functions dynamically derive information to constrain subsequent processing steps. The derived information allows in particular the application of an existing index or some other traditional join strategy. After demonstrating this technique on various examples, we present the results of a practical performance evaluation.