进入Internet of X年代，RDF流数据处理已经成为大多数互联系统的基础操作，面向RDF流数据推理由于其高复杂度的特征和高实时性要求之间的矛盾，基于CPU的体系结构难以胜任该类工作。本项目采用CPU/GPU混合架构实现细粒度并行RDF流数据处理方法，主要研究内容有：1）基于CQELS设计出一种并行的处理大规模RDF流的查询语言CQELS+，并从理论上验证其可并行性。同时，针对RDF流汇聚带来的信息不完整、语义冲突等问题，研究相应的策略和算法。2）在充分了解RDF流查询语言CQELS+特性的基础上，研究面向GPU并行处理的最低计算代价表达和存储结构，在查询模式的分解与合并和GPU算法效率寻求平衡。3）基于异步消息的混合系统架构，实现基于GPU加速的RDF流处理并行/分布式处理架构，并构建原型系统。本研究成果可广泛用于物联网、社交网络和智慧城市等系统的语义流数据处理。

In the Internet of X's, the RDF stream data processing has become the basic operation of the most interconnected system. For high complex features of RDF stream and high real-time requirements of its reasoning, CPU based architecture is not suitable for RDF stream reasoning. In this project, CPU/GPU hybrid architecture based fine-grained parallel framework will be studied to process RDF stream data, the research contents as follow: 1).A large-scale RDF query language CQELS+ will be design based on CQELS, and its parallelism will be verified. Incomplete information and semantic conflict brought by RDF flow convergence also will be treated with strategies and corresponding algorithms. 2) On the foundation of full understanding of the designed RDF stream query language CQELS+, we will find the GPU oriented minimum cost expression and storage structure for RDF stream parallel processing, and seek a balance between and decomposition and merging of query mode and GPU algorithm efficiency. 3) Based on the hybrid system architecture of asynchronous message, the GPU based parallel distributed processing architecture of RDF stream processing will is realized to build a prototype system. The results of this research can be widely used in the data processing of the system of Internet of things, social network and intelligent city.