Optimizing Enterprise-Scale OWL 2 RL Reasoning in a Relational Database System

OWL 2 RL was standardized as a less expressive but scalable subset of OWL 2 that allows a forward-chaining implementation. However, building an enterprise-scale forward-chaining based inference engine that can 1) take advantage of modern multi-core computer architectures, and 2) efficiently update inference for additions remains a challenge. In this paper, we present an OWL 2 RL inference engine implemented inside the Oracle database system, using novel techniques for parallel processing that can readily scale on multi-core machines and clusters. Additionally, we have added support for efficient incremental maintenance of the inferred graph after triple additions. Finally, to handle the increasing number of owl:sameAs relationships present in Semantic Web datasets, we have provided a hybrid in-memory/disk based approach to efficiently compute compact equivalence closures. We have done extensive testing to evaluate these new techniques; the test results demonstrate that our inference engine is capable of performing efficient inference over ontologies with billions of triples using a modest hardware configuration.