High Performance Graph Computing System Based On CPU-Phi Heterogeneous Architecture

With the advent of large data age,technological changes associated with large data have become a hot topic in academia and industry.Because of the abundant expression ability of the graphs in the description of the relationship between the objects,the graphs are widely used in many areas such as planning of traffic routes,citation relations of paper works,security event analysis,social network analysis,biomedicine and so on.With the increasing scale of graph data,distributed large graph processing technology has been developed rapidly.These technologies include distributed graph computa-tion model,large graph data storage and management,various high performance graph algorithm,high performance graph computing system and so on.Intel Xeon Phi is an Intel coprocessor for high-performance computing.Phi ex-tends vector computing capacity on the basis of maintaining the x86 architecture,with the advantages of computing power and ease of development.Cgraph is the only graph computing system implemented on CPU-Phi heterogeneous systems,but Cgraph has some shortcomings in system implementation.Such as,the vector computing ability of Phi is not fully utilized,the message buffer mechanism is inefficient and thread man-agement is not effective,and Cgraph also did not achieve on-line data segmentation.In the aspect of graph segmentation,most of the graph computing systems do not take into account the difference in the capacity of the compute nodes when dividing the graph data,so that the computational power of the heterogeneous computing nodes can not be fully utilized.This paper studies the high performance graph computing system on CPU-Phi het-erogeneous platform,and makes a useful exploration for the full use of Phi architecture to optimize the graph computing.The main work of the paper includes the following three aspects:(1)Optimize the graph computing process by using the architecture of Phi;(2)Design the architecture-aware graph segmentation algorithm for different CPU and Phi processing capabilities;(3)Based on the above technology,a stand-alone layout graph computing system Pgraph is implemented on the CPU-Phi heterogeneous archi-tecture and compared with Cgraph.At the same time,the validity of Pgraph in design and implementation is verified by experiments.