SDN中一种多控制器负载均衡方案

A Multi-Controller Load Balancing Scheme in SDN

针对软件定义网络(software defined network,SDN)中控制器子域分区不合理导致的控制器负载分配不平衡、网络通信性能下降的问题,本文提出一种基于负载均衡的多控制器部署(multi-controller load balancing,MLB)方案.在初始静态网络中,提出一种基于域内和域间通信成本的负载平衡控制器部署模型,并将流量请求转换为排队模型.MLB将近邻传播(affinity propagation,AP)算法中的偏置参数和阻尼系数这两个参数作为粒子,通过粒子群优化(particle swarm optimization,PSO)算法对其进行智能调整,解决集群性能受偏置参数和阻尼系数初始值影响的问题,得到合理的网络规划.实验结果表明:与亲缘传播算法和遗传算法(genetic algorithm,GA)相比,该方案可以提供更稳定、准确和负载均衡的多控制器部署.

Aiming the problem of unbalanced controller loads distribution and network communication performance degradation caused by unreasonable subdomain partitioning of controllers in software defined network(SDN),a multi-controller load balancing scheme is proposed.In the initial static network,a load balancing controller deployment model based on the intra-domain and inter-domain communication cost has been proposed,and the traffic requests are transformed into a queuing model.MLB takes the two parameters that bias parameters and convergence coefficients in affinity propagation algorithm as particles,then adjusts it intelligently by particle swarm optimization(PSO)algorithm,to solve the problem of clustering performance being affected by the initial values of the bias parameters and convergence coefficients,getting the reasonable network planning.Experimental results show that compared with affinity propagation algorithm and genetic algorithm,this scheme can provide more stable,accurate and load balancing multi-controller deployment.