Cooperative Query Processing On Heterogeneous Processors

As a markup language for describing Web resources,RDF(Resource Description Framework)is usually used to represent graphical data due to its simplicity and flexibility.SPARQL(Simple Protocol and RDF Query Language)is a standard RDF query language recommended by W3C(World Wide Web Consortium).With the rapid growth of RDF data,how to efficiently respond to SPARQL queries has become one of the major challenges in the current status of RDF data processing.Many existing SPARQL query systems based on CPU(Central Processing Unit)have encountered performance bottlenecks to some extent.In recent years,GPU(Graphic Processing Unit)has brought the possibility of improving SPARQL query processing performance due to its powerful computing capabilities.The Grace system was designed and implemented to provide efficient SPARQL query response for heterogeneous processor co-processing mode to fully exploit the computational performance of GPU processors.The system contains multi-concurrent pipeline query plan generation method,the generated query plan is easy to be executed in parallel and can reduce the delay of data transfer between heterogeneous processors.Multiple-concurrent query operators are used to parallelize the data scanning and connection.In order to minimize performance loss caused by data transmission and synchronization,the system optimizes data transmission and communication between heterogeneous processors.Experiments on LUBM(Lehigh University Benchmark)and WATDIV(Waterloo SPARQL Diversity Test Suite)datasets show that the average query performance of Grace is about 13 x and 3x better under cold cache conditions compared to the traditional popular SPARQL query engines RDF-3X and Triple Bit,and about 7x and 2x better under warm cache conditions.The average query performance under GPU acceleration is improved by about 10% to 50%.In addition,the Grace system exhibits good scalability regarding work threads.Compared to the single-threaded mode,the average speedup of query performance is about 4 times.