BitMat – Scalable Indexing and Querying of Large RDF Graphs

The growing size of Semantic Web data expressed in the form of Resource Description Framework (RDF) has made it necessary to develop effective ways of storing this data to save space and to query it in a scalable manner. SPARQL – the query language for RDF data – closely follows SQL syntax. As a natural consequence most of the RDF storage and querying engines are based on modern database storage and query optimization techniques. Previous work has tried to use vertical partitioning using column stores (C-Store, MonetDB) and 6-way indexing (RDF-3X, Hexastore) for storage and querying of RDF data. Although these approaches perform well for highly selective queries, for queries having low-selectivity triple patterns, scalability of the querying method and optimizations still remain a challenge. In this paper we present a new way of storing RDF graphs in run-length-encoded bit-vector format called BitMat, and we propose a novel two-phase SPARQL join query processing algorithm. In the first phase it prunes the candidate RDF triples, and in the next phase, it stitches the pruned RDF triples together to generate final results. Our query processing method does not build intermediate join tables and works directly on the compressed data. Our evaluation shows that BitMat not only provides an efficient method of storage of the RDF graphs, but our join query processing algorithm scales well for low-selectivity join queries, where state-of-the-art RDF query processors face problems.