BitMat : An In-core RDF Graph Store for Join Query Processing

With the growing size of RDF data sources, the need for a compact representation providing efficient query interface has become compelling. In this paper, we introduce BitMat, a main memory based compressed bit-matrix structure. The key aspects of BitMat are as follows: i) its RDF graph representation is very compact compared to the conventional disk-based and existing main-memory RDF stores, and ii) SPARQL Basic Graph Pattern (BGP) queries are processed using a novel method employing initial pruning technique, followed by a subgraph matching algorithm on an BitMat, without uncompressing it at any point. This enables processing queries on very large RDF graphs completely in-memory. We present the key concepts of the BitMat structure and our BGP query processing algorithm. Finally we compare memory utilization and query performance of BitMat with contemporary triplestores for up to 51 million triples.