基于交换机迁移的SDN控制平面负载均衡研究

发布时间：2018-04-12 09:31

本文选题：软件定义网络 + 负载均衡　；参考：《安徽大学》2017年硕士论文

【摘要】：近年来关于软件定义网络(Software Defined Network,SDN)的研究已成为未来网络的热点课题。控制平面负载均衡研究作为SDN中一项基础而又重要的课题,旨在有效提高控制平面可扩展性。研究发现控制器与交换机之间的静态映射关系是导致控制平面负载失衡的关键因素。目前大多数算法采用将过载控制器下的交换机迁移到其他控制器的方法来解决控制平面负载失衡问题。本文针对缺乏灵活的负载均衡架构,提出了基于三层结构模型的控制平面负载均衡算法。同时针对交换机迁移方式过于简单、在迁移过程中没有考虑通信开销的问题,提出了基于多目标优化的控制平面负载均衡算法。本文的主要研究工作如下:(1)针对缺乏灵活的负载均衡架构,提出了基于控制器管理层-控制器集群-数据层三层结构模式的负载均衡架构MCD(controller Management-Controller cluster-Data layer)。MCD 在控制器集群上部署控制器管理层,控制器管理层主要负责监测控制器负载,并根据负载信息做出相应决策和动作指令。MCD能够有效减少控制器之间的通信开销,使控制器专注于制定数据分组转发决策,在一定程度上提高了控制器的性能。在MCD的基础上,针对现有交换机迁移方式仍存在各种缺陷,提出了控制平面负载均衡算法SMLB(Switch Migrate for Load Balancing)。该算法站在全局角度判断是否进行负载均衡以及何时进行负载均衡,避免了频繁的负载均衡带来不必要的通信开销和网络延迟,并且综合考虑负载因素和距离因素对均衡效果的影响。仿真结果表明,该算法可以有效均衡控制平面负载,降低网络延迟,提高控制平面可拓展性。(2)本文针对交换机迁移方式过于简单和迁移过程中未考虑通信开销的问题,提出了基于多目标优化的控制平面负载均衡算法M-DSMA(Dynamic Switch Migration Algorithm based on Multi-objective)。首先将交换机与控制器之间的映射关系转变为0-1矩阵优化问题;其次利用NSGA-II算法同时优化控制平面负载均衡度和交换机迁移所产生的通信开销这两个相互冲突的指标,从而得到合适的交换机与控制器部署方案。该算法利用组编码方式将交换机与控制器之间的映射关系编码为遗传学中的个体基因,将控制平面的负载均衡度和交换机迁移开销定义为个体的适应度函数,依据种群进化得到的最优解迁移交换机,均衡控制平面负载。仿真结果表明,M-DSMA可以有效均衡控制平面负载,降低通信开销,提高控制平面可扩展性,具有较高的实际应用价值。
[Abstract]:In recent years, the research on Software Defined Network (SDN) has become a hot topic in the future.As a basic and important topic in SDN, load balancing in control plane is aimed at improving the scalability of control plane.It is found that the static mapping relationship between the controller and the switch is the key factor leading to the control plane load imbalance.At present, most algorithms adopt the method of migrating switch under overload controller to other controllers to solve the problem of control plane load imbalance.In this paper, a control plane load balancing algorithm based on three-layer structure model is proposed for the lack of flexible load balancing architecture.At the same time, a load balancing algorithm based on multi-objective optimization is proposed to solve the problem that the switching mode is too simple and the communication overhead is not considered in the migration process.The main research work of this paper is as follows: (1) aiming at the lack of flexible load balancing architecture,This paper presents a load-balancing architecture, MCD(controller Management-Controller cluster-Data layer).MCD, based on controller management-controller cluster-data layer structure, which is mainly responsible for monitoring controller load.According to the load information, the corresponding decision and action instruction .MCD can effectively reduce the communication overhead between the controllers, make the controller focus on making the decision of data packet forwarding, and improve the performance of the controller to a certain extent.On the basis of MCD, a new load balancing algorithm, SMLB(Switch Migrate for Load balancing algorithm, is proposed to overcome the defects of the existing switch migration mode.The algorithm is based on the global perspective to determine whether and when load balancing takes place, thus avoiding unnecessary communication overhead and network delay caused by frequent load balancing.And the influence of load factor and distance factor on the balance effect is considered synthetically.The simulation results show that the algorithm can effectively balance the load of the control plane, reduce the network delay, and improve the scalability of the control plane. In this paper, the transfer mode of the switch is too simple and the communication overhead is not considered in the migration process.A control plane load balancing algorithm, M-DSMA(Dynamic Switch Migration Algorithm based on Multi-objective, is proposed based on multi-objective optimization.Firstly, the mapping relationship between switch and controller is transformed into a 0-1 matrix optimization problem, and then NSGA-II algorithm is used to optimize the two conflicting indexes of control plane load balance degree and communication overhead caused by switch migration.Thus the appropriate switch and controller deployment scheme is obtained.In this algorithm, the mapping between switch and controller is encoded as individual gene in genetics by group coding, and the load balance and switch transfer overhead in control plane are defined as fitness function of individual.The optimal solution migration switch based on population evolution is used to balance the plane load.The simulation results show that M-DSMA can effectively balance the load of the control plane, reduce the communication overhead and improve the expansibility of the control plane.
【学位授予单位】：安徽大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.02

【相似文献】