Intelligent SPARQL Endpoints: Optimizing Execution Performance by Automatic Query Relaxation and Queue Scheduling

The Web of Data is widely considered as one of the major global repositories populated with countless interconnected and structured data prompting these linked datasets to be continuously and sharply increasing. In this context the so-called SPARQL Protocol and RDF Query Language is commonly used to retrieve and manage stored data by means of SPARQL endpoints, a query processing service especially designed to get access to these databases. Nevertheless, due to the large amount of data tackled by such endpoints and their structural complexity, these services usually suffer from severe performance issues, including inadmissible processing times. This work aims at overcoming this noted inefficiency by designing a distributed parallel system architecture that improves the performance of SPARQL endpoints by incorporating two functionalities: (1) a queuing system to avoid bottlenecks during the execution of SPARQL queries; and (2) an intelligent relaxation of the queries submitted to the endpoint at hand whenever the relaxation itself and the consequently lowered complexity of the query are beneficial for the overall performance of the system. To this end the system relies on a two-fold optimization criterion: the minimization of the query running time, as predicted by a supervised learning model; and the maximization of the quality of the results of the query as quantified by a measure of similarity. These two conflicting optimization criteria are efficiently balanced by two bi-objective heuristic algorithms sequentially executed over groups of SPARQL queries. The approach is validated on a prototype and several experiments that evince the applicability of the proposed scheme.