Parallel Tractability of Ontology Materialization: Technique and Practice

Abstract

Materialization is an important reasoning service for many ontology-based applications, but the rapid growth of semantic data poses the challenge to efficiently perform materialization on largescale ontologies. Parallel materialization algorithms work well for some ontologies, although the reasoning problem for the used ontology language is not in NC, i.e., the theoretical complexity class for parallel tractability. This motivates us to study the problem of parallel tractability of ontology materialization from a theoretical perspective. We focus on the datalog rewritable ontology languages DL-Lite and Description Horn Logic (DHL) and propose algorithms, called NC algorithms, to identify classes of ontologies for which materialization is tractable in parallel. To verify the practical usability of the above results, we analyze different benchmarks and realworld datasets, including LUBM and the YAGO ontology, and show that for many ontologies expressed in DHL materialization is tractable in parallel. The implementation of our optimized parallel algorithm shows performance improvements over the highly optimized state-of-the-art reasoner RDFox on ontologies for which materialization is tractable in parallel.