Graph Computing Systems Based on CPU-Phi Heterogeneous Platform
Wang Jian;Hua Bei;School of Computer Science and Technology, University of Science and Technology of China;
Graph computing has a wide range of applications in the era of big data, and high performance graph computing system has become one of the important platforms for information processing. With the rapid increase in the computing power of manycore processors, multicore/manycore collaboration has become one of the important ways to build high performance computing systems. The Intel Xeon Phi is a manycore processor compatible with X86 architectures, and has high parallel compuing power and memory bandwidth. However, there are few graph computing systems built on a CPU-Phi heterogeneous platform, apart from Cgraph-the only known such system in the literature. Aiming to overcome the shortcomings of Cgraph, this paper redesigns a graph computing system called Pgraph on a server with a multicore CPU and a Phi coprocessor. Pgraph mainly optimizes the graph iterative processing, which includes a message buffer design that has high parallel degree and memory access efficiency, a fully parallelization and vectorization scheme of iterative processing, and dynamic task assignment and thread scheduling based on load balancing. Pgraph system outperforms Cgraph in all the experiments carried in this paper.
【CateGory Index】： TP332