A Data Partitioning Approach for Parallelizing Rule Based Inferencing for Materialized OWL Knowledge Bases

Materialized knowledge bases perform inferencing when data is loaded into them, so that answering queries is reduced to simple lookup and thus are faster. A major bottleneck of such a system is the inferencing process, which is slow and memory intensive. In this work we examine the problem of speeding up as well as scaling the inferencing process for OWL Knowledge Bases, that employ rule based reasoners. We propose a data partitioning approach for this problem, in which the input data is partitioned into smaller chunks that are then processed independently. We propose a parallel reasoning algorithm and show the correctness of our technique for the class of rule-sets obtained from OWL ontologies. We present two optimizations of this algorithm to further improve the performance. Finally, we present an implementation based on a popular open source tool. In our experiments using a standard OWL benchmark we have observed speedups of upto 18x, on a parallel cluster of 16 processors.