SRDF_QDAG: An efficient end-to-end RDF data management when graph exploration meets spatial processing

Abstract

The popularity of RDF has led to the creation of several datasets (e.g., Yago, DBPedia) with different natures (graph, temporal, spatial). Different extensions have also been proposed for SPARQL language to provide appropriate processing. The best known is GeoSparql, that allows the integration of a set of spatial operators. In this paper, we propose new strategies to support such operators within a particular TripleStore, named RDF QDAG, that relies on graph fragmentation and exploration and guarantees a good compromise between scalability and performance. Our proposal covers the different TripleStore components (Storage, evaluation, optimization). We evaluated our proposal using spatial queries with real RDF data, and we also compared performance with the latest version of a popular commercial TripleStore. The first results demonstrate the relevance of our proposal and how to achieve an average gain of performance of 28% by choosing the right evaluation strategies to use. Based on these results, we proposed to extend the RDF QDAG optimizer to dynamically select the evaluation strategy to use depending on the query. Then, we show also that our proposal yields the best strategy for most queries.