Parallel Data Access for Multiway Rank Joins

Rank join operators perform a relational join among two or more relations, assign numeric scores to the join results based on the given scoring function and return K join results with the highest scores. The top-K join results are obtained by accessing a subset of data from the input relations. This paper addresses the problem of getting topK join results from two or more search services which can be accessed in parallel, and are characterized by non negligible response times. The objectives are: i) minimize the time to get top-K join results. ii) avoid the access to the data that does not contribute to the top-K join results. This paper proposes a multi-way rank join operator that achieves the above mentioned objectives by using a score guided data pulling strategy. This strategy minimizes the time to get top-K join results by extracting data in parallel from all Web services, while it also avoids accessing the data that is not useful to compute top-K join results, by pausing and resuming the data access from different Web services adaptively, based on the observed score values of the retrieved tuples. An extensive experimental study evaluates the performance of the proposed approach and shows that it minimizes the time to get top-K join results, while incurring few extra data accesses, as compared to the state of the art rank join operators.