Practical parallel string matching framework for RDF entailments with GPUs

摘要

Resource Description Framework (RDF) is a commonly used format for semantic web processing. It basically contains strings representing items and their relationships which can be queried or inferred. In this paper, we propose a framework for processing large RDF data sets. It is based on Brute-force string matching on GPUs (BFG). Graphics Processing Units (GPUs) are used as a parallel platform that allows thousands of threads to find RDF data. Our search algorithm is customized to suit the nature of RDF processing and GPU memory architecture. Then, the algorithm is integrated into the proposed framework for computing queries and chaining rules for RDF data. Experiments show that utilizing these algorithms can achieve the speedup of 7 times for querying and for forward chaining compared to using the sequential version. The proposed framework can achieve a string comparison rate of 67,000 comparisons per second using 2 GPUs.