Graph Pattern Matching

Abstract

This paper presents an empirical study of how a wide spectrum of systems handle the graph pattern matching problem. Our approach is to take the well-known LUBM benchmark, model it across various domains (relational, RDF, property graph), and execute the benchmark queries on the corresponding systems. We evaluate the systems using a large data instance on a single machine (the largest dataset is LUBM-8000, which contains over 1 billion RDF triples). Additionally, we provide a brief analysis of how different cases of graph pattern matching problem are stressed by the benchmark queries. Our main finding is that, contrary to popular belief and various vendors' claims, modern native graph stores do not necessarily offer a competitive advantage over traditional relational and RDF stores, even for the graph-specific problem of pattern matching. To the best of our knowledge, this is the first independent empirical comparison of different approaches towards pattern matching performed on a large scale graph.