Informatics and Computer Science ~ - HOUAND Performance of Recursive Query Processing Methods in the Presence of Cycles ATAKAN KURT and MERAL OZSOYOGLU Department of Computer Engineering and Science, Case Western Reser~'e UniL~ersity, Cleveland, 0 t t 44106, USA Communicated by Ben Wah ABSTRACT An experimental performance analysis of three recursive query processing methods [the magic-sets method (MAG), the magic-counting method (MCS), and the synchro- nized-counting method (SCM)] in deductive databases in the presence of cycles is presented. We define a set of characteristics that affects the performance of query processing methods for cyclic data. Although there are other experimental performance analysis studies of recursive query processing methods in the literature, these studies have covered only acyclic databases. There are only some bounds for the worst case and best case analysis of recursive query processing methods on cyclic data. This paper introduces a set of graph characteristics on which similar performance studies can be based, and it is, to our knowledge, the first comparative experimental study on the performance of recursive query processing methods for cyclic data. 1. INTRODUCTION Deductive databases are based on the idea of combining logic program- ming and relational databases. We consider a deductive database in which Horn clauses without function predicates are used to express inference logic. Programs are defined in a specific formalism, called Datalog [14], a Prolog-like logic programming language. While a top-down approach is used in Prolog to process a query, bottom-up execution techniques are used in Datalog. The program execution in Prolog is of exponential INFORMATION SCIENCES 91, 147-192 (1996) © Elsevier Science Inc. 1996 0020-0255/96/$15.00 655 Avenue of the Americas, New York, NY 1001ll PII S0020-0255(96)00021-7