电子学报 ›› 2016, Vol. 44 ›› Issue (9): 2197-2202.DOI: 10.3969/j.issn.0372-2112.2016.09.026

• 学术论文 • 上一篇    下一篇

基于离散粒子群算法的数据中心网络流量调度研究

林智华1,2, 高文2, 吴春明2, 李勇燕3   

  1. 1. 福建江夏学院电子信息科学学院, 福建福州 350108;
    2. 浙江大学计算机学院, 浙江杭州 310027;
    3. 绍兴文理学院上虞分院, 浙江上虞 312300
  • 收稿日期:2014-11-10 修回日期:2015-03-06 出版日期:2016-09-25 发布日期:2016-09-25
  • 作者简介:林智华 男,1972年12月生于福建福州,现为福建江夏学院电子信息科学学院副教授,主要研究方向为新一代网络技术.E-mail:lindiva@126.com
  • 基金资助:

    973计划(No.2012CB315903);浙江省重点科技创新团队(No.2011R50010-21);国家科技支撑计划(No.2014BAH24F01);国家自然科学基金(No.61379118)

Data Center Network Flow Scheduling Based on DPSO Algorithm

LIN Zhi-hua1,2, GAO Wen2, WU Chun-ming2, LI Yong-yan3   

  1. 1. College of Electronics Information Engineering, Fujian Jiangxia University, Fuzhou, Fujian 350108, China;
    2. College of Computer Science, Zhejiang University, Hangzhou, Zhejiang 310027, China;
    3. University of Shaoxing at Shangyu, Shangyu, Zhejiang 312300, China
  • Received:2014-11-10 Revised:2015-03-06 Online:2016-09-25 Published:2016-09-25

摘要:

数据中心网络利用多个并行路径为集群计算等网络服务提供高对分带宽.然而,现有的流量调度算法可能会引起链路负载不均衡,核心交换机冲突加剧,造成网络总体性能降低.本文将流调度问题转化成0-K背包问题求解,提出基于离散粒子群的流调度算法DPSOFS(Discrete Particle Swarm Optimization Flow Scheduling).该算法根据Fat-Tree结构特点定义了粒子速度、位置和运算规则,以两次迭代冲突流个数差值作为目标函数,并限定路径搜索范围,减少随机搜索的盲目性.仿真实验验证了该算法对减少流冲突快速有效,能提高网络对分带宽.

关键词: Fat-Tree, 数据中心网络, 离散粒子群, 流调度

Abstract:

Data center networks leverage multiple parallel paths connecting end host pairs to offer high bisection bandwidth forcluster computing applications.However,state of the art flow scheduling algorithms may cause unfair link utilization and saturation of core switches,resulting in overall bandwidth loss.In the paper,we regard the flow scheduling problem as a 0-K knapsack problem and propose a new flow scheduling algorithm named DPSOFS based on DPSO.DPSOFS formulates the position,velocity and their operation rules of particles according to Fat-Tree topology structure,and defines objective function as the difference of the number of conflict flows between two iterations.Moreover,our proposed mechanism reduces random search blindness by limiting the range of the path search.The simulation suggests that it can improve overall network bisection efficiently.

Key words: Fat-Tree, data center network, DPSO, flow scheduling

中图分类号: