UPSP: Unique predicate-based source selection for SPARQL endpoint federation

Abstract

Efficient source selection is one of the most important optimization steps in federated SPARQL query processing as it leads to more efficient query execution plan generation. An over-estimation of the data sources will generate extra network traffic by retrieving irrelevant intermediate results. Such intermediate results will be excluded after performing joins between triple patterns. Consequently an over-estimation of sources may result in increased query execution time. Devising triple patterns join-aware source selection approaches has shown to yield great improvement potential. In this work, we present UPSP, a new source selection approach for SPARQL query federation over multiple SPARQL endpoints. UPSP makes use of the subject-subject, subject-object, object-subject, and object-object joins information stored in an index structure to perform efficient triple patterns join-aware source selection. Our evaluation results on FedBench shows that UPSP outperforms state-of-the-art source selection approaches by selecting smaller number of sources (without losing recall) and reducing the query execution times.