Towards efficient distributed SPARQL queries on linked data

The fast growth of the web of linked data raises new challenges for distributed query processing. Different from traditional federated databases, linked data sources cannot cooperate with each other. Hence, sophisticated optimization techniques are necessary for efficient query processing. Source selection and distributed join operations are key factors concerning performance of linked data query engines. In this paper, we propose identifier graph based source selection taking into account the logical relationship between triple patterns, and develop effective solutions for distributed join operations to avoid program errors and to minimize network traffic. In experiments, we demonstrate the practicability and efficiency of our approaches on a set of real-world queries and data sources from the Linked Open Data cloud. With the implemented prototype system, we achieve a significant improvement in the accuracy of source selection and query performance over state-of-the-art federated query engines.