Investigations on path indexing for graph databases

Graph databases have become an increasingly popular choice for the management of the massive network data sets arising in many contemporary applications. We investigate the effectiveness of path indexing for accelerating query processing in graph database systems, using as an exemplar the widely used open-source Neo4j graph database. We present a novel path index design which supports efficient ordered access to paths in a graph dataset. Our index is fully persistent and designed for external memory storage and retrieval. We also describe a compression scheme that exploits the limited differences between consecutive keys in the index, as well as a workload-driven approach to indexing. We demonstrate empirically the speed-ups achieved by our implementation, showing that the path index yields query run-times from 2x up to 8000x faster than Neo4j. Empirical evaluation also shows that our scheme leads to smaller indexes than using general-purpose LZ4 compression. The complete stand-alone implementation of our index, as well as supporting tooling such as a bulk-loader, are provided as open source for further research and development.