基于离散粒子群算法的数据中心网络流量调度研究  

Data Center Network Flow Scheduling Based on DPSO Algorithm

作　　者：林智华[1,2] 高文[2] 吴春明[2] 李勇燕[3] 

机构地区：[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

出　　处：《电子学报》2016年第9期2197-2202,共6页Acta Electronica Sinica

基　　金：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)

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.Data center networks leverage multiple parallel paths connecting end host pairs to offer high bisection bandwidth forcluster computing applications. However,state of the art flowscheduling algorithms may cause unfair link utilization and saturation of core switches,resulting in overall bandwidth loss. In the paper,we regard the flowscheduling problem as a 0-K knapsack problem and propose a newflowscheduling 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.

关 键 词：Fat-Tree 数据中心网络 离散粒子群 流调度 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]