软件定义网络中面向服务的负载均衡机制

发布时间：2018-03-16 10:29

本文选题：SDN　切入点：分布式控制器　出处：《重庆邮电大学》2016年硕士论文　论文类型：学位论文

【摘要】：随着移动互联网、云计算与大数据等服务的兴起和发展,网络规模和数据流量成指数级增长,同时由于互联网新型应用的不断丰富,用户服务质量(QoS)需求发生了巨大改变。因此,如何均衡网络流量、保证QoS等成为当前网络亟待解决的问题,而传统互联网已经越来越难以满足当前发展需求。软件定义网络(Soft-Defined Networking,SDN)作为新型互联网架构,将数据平面与控制平面分离,简化了网络管理的复杂性,具有全局网络状态视图,能够灵活地实现网络流量控制和QoS保障,但同样面临可扩展性问题。因此,针对网络负载不均衡以及控制平面可扩展性问题,提出软件定义网络中面向服务的负载均衡机制,主要研究工作包括:第一,提出了一种分布式控制器负载均衡模型,将交换机对间的流请求信息作为控制器管理的基本单元,控制器周期性地发布各自的流请求信息数,并引入流请求偏离均值数用以控制器感知自身负载状态。在此基础上,进一步提出了基于负载感知的负载均衡算法,该算法采用流请求信息分配策略,考虑空闲控制器的当前负载和传播时延,将过载控制器上的部分流请求信息分配给流请求偏离均值数和传播时延均小的空闲控制器。同时为了避免过载控制器同时执行负载均衡算法造成的网络状态不一致性,每个控制器都维护一张流请求偏离均值表,控制器根据表中偏离均值数的大小顺序执行负载均衡算法。实验结果表明,该模型能够快速有效地调整控制器负载,具有较低的时间复杂度和更好的鲁棒性。第二,设计了服务感知的自适应链路负载均衡机制,该机制通过控制器提供扩展的北向接口感知网络中的服务类型并周期性地监测网络状态,给出了自适应链路负载均衡算法。该算法引入基于QoS感知的链路权重,该权重通过控制器获取的实时QoS参数来衡量链路的综合质量,从而为服务选择当前链路质量最优的路径进行数据转发,降低网络负载分布的不均衡性。并针对不同的服务类型,进一步提出动态QoS路由优化策略,采用拉格朗日松弛技术为服务计算一条满足QoS约束的路径。实验结果表明,该机制能够有效地均衡网络流量,平均链路带宽利用率最高达到79%,并实现了服务的QoS保证。
[Abstract]:With the rise and development of mobile Internet, cloud computing, big data and other services, the network scale and data traffic are increasing exponentially. Therefore, how to balance the network traffic and ensure the QoS becomes an urgent problem in the current network. The traditional Internet has become more and more difficult to meet the current development needs. As a new type of Internet architecture, software defined network Soft-Defined networking (SDN) separates the data plane from the control plane, simplifies the complexity of network management, and has a global network state view. It can flexibly realize network traffic control and QoS guarantee, but it also faces the problem of scalability. Therefore, aiming at the problem of network load imbalance and control plane scalability, this paper proposes a service-oriented load balancing mechanism defined by software in the network. The main research work includes: first, a distributed controller load balancing model is proposed. The flow request information between switch pairs is taken as the basic unit of controller management, and the controller periodically issues their respective flow request information. Based on this, a load balancing algorithm based on load awareness is proposed, which adopts flow request information allocation strategy. Considering the current load and propagation delay of the idle controller, The partial flow request information on the overload controller is assigned to the idle controller where the flow request deviates from the mean number and the propagation time delay is small. Meanwhile, in order to avoid the network state inconsistency caused by the overload controller executing load balancing algorithm simultaneously, Each controller maintains a flow request deviation mean table, and the controller performs a load balancing algorithm according to the order of deviation from the mean number in the table. The experimental results show that the model can adjust the controller load quickly and effectively. It has lower time complexity and better robustness. Secondly, a service-aware adaptive link load balancing mechanism is designed. This mechanism provides an adaptive link load balancing algorithm through the controller to provide extended northward interface sensing network service type and periodically monitor the network state. The algorithm introduces link weight based on QoS perception. The weight measures the overall quality of the link through the real-time QoS parameters obtained by the controller, and then selects the path with the best link quality for the service to transmit data, reduces the imbalance of the network load distribution, and aims at different service types. Furthermore, a dynamic QoS routing optimization strategy is proposed. Lagrangian relaxation technique is used to calculate a path satisfying QoS constraints. Experimental results show that the proposed scheme can effectively balance network traffic. The average link bandwidth utilization is up to 79, and the QoS guarantee of service is realized.
【学位授予单位】：重庆邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP393.09

【参考文献】