Indexed Step Wise Semi Naive Evaluation for Recursive Queries

A nice property of the semi naive evaluation of recursive rules is that it does not repeat rule rings in di erent iterations In a recent study W however it was observed that intermediate steps during a rule ring may be repeated in di erent iterations and a notion of the step wise semi naive property and an evaluation with this property were proposed to capture the semi naive prop erty at intermediate steps of a rule ring In this paper we further extend the semi naive notion to the build up of index for computing relations that grow monotonically in each iteration Hash tables are chosen as the index in this study but the idea can be applied to other types of index in general A modi cation of the step wise semi naive evaluation based on the semi naive build up of hash tables is proposed In most cases join set di erence and set union operations performed to re a rule are implemented in a single I O scan of relations that are cumulated up to that iteration Experiments conducted on a few typical recursive queries and test data show that the proposed evaluation is much more I O cost e ective than both the semi naive evaluation and the step wise semi naive evaluation