基于OpenFlow的负载均衡的改进与优化

发布时间：2018-11-14 08:28

【摘要】：随着大数据、云计算以及网络虚拟化技术的飞速发展,网络中的业务流量急剧增长,为了避免网络拥塞,提升网络系统的容错性以及鲁棒性,对于负载均衡技术的研究变得至关重要。负载均衡技术可以在一定程度上改善网络节点流最分配不均,部分节点过载,而其他节点处于空闲状态的情况。由于传统网络架构网络配置复杂,实施负载均衡灵活性差,而且无法充分利用全局网络资源进行流量调度。因此,亟待提出一种新型网络架构进行革新。基于OpenFlow协议的软件定义网络,其核心理念就是控制平面与转发平面的分离。SDN架构打破了传统网络的层次化概念,以扁平化的方式实现全局的网络控制。OpenFlow控制器位于控制平面,主要实现网络的集中化管理,制定转发策略。其向上提供API供应用程序调用,向下通过标准的南向接口协议下发流表。OpenFlow交换机位于转发平面,主要根据控制器下发的转发决策机制进行高速数据转发。控制平面和转发平面的解耦,可以灵活把控全局网络,提升负载均衡的有效性。为此,本文提出了一种基于OpenFlow的负载均衡改进方案。本文针对数据中心中典型的胖树网络拓扑结构,首先确定影响负载均衡的两个关键问题:业务请求分配和寻找最短路径。然后针对这两个问题进行改进和优化,进而提出新的负载均衡方案。首先根据全局网络拓扑信息以及服务器负载情况,设计出一种通配符规则AWR对业务请求进行分配,进而获取到请求主机和处理服务器之间的配对信息。该规则的核心思想即根据服务器权重优先级进行请求聚合,实现流规则最小化。然后,根据得到的配对信息,提出一种基于原始Dijkstra算法的改进算法AD,对多条路径进行选路,找出最短路径。该算法改进的关键在于综合考虑路径瓶颈带宽及权重等信息来作出选路策略,进而实现带宽利用率和时延的优化。本文最后采用在虚拟实验平台Mininet上搭建Fat-Tree拓扑结构,分别针对多种不同场景进行测试,得出实验相关数据。然后根据带宽利用率和数据包往返时延两项性能指标,验证本文提出的负载均衡方案的优越性。
[Abstract]:With the rapid development of big data, cloud computing and network virtualization technology, traffic in the network has increased dramatically. In order to avoid network congestion and improve the fault tolerance and robustness of network system, It is very important for the research of load balancing technology. Load balancing technology can improve the situation that the network node flow is most unevenly distributed, some nodes are overloaded, and other nodes are idle. Because of the complexity of network configuration in traditional network architecture, the flexibility of load balancing is poor, and the global network resources can not be fully utilized for traffic scheduling. Therefore, it is urgent to propose a new network architecture for innovation. The core idea of the software definition network based on OpenFlow protocol is to separate the control plane from the forwarding plane. The SDN architecture breaks the hierarchical concept of traditional network. The OpenFlow controller is located in the control plane, which mainly realizes the centralized management of the network and formulates the forwarding strategy. The OpenFlow switch is located in the forwarding plane, which is mainly based on the forwarding decision mechanism of the controller to carry out high-speed data forwarding. The decoupling of control plane and forwarding plane can flexibly control the global network and enhance the effectiveness of load balancing. Therefore, an improved load balancing scheme based on OpenFlow is proposed in this paper. In this paper, two key problems affecting load balancing are first identified for the typical fat tree network topology in the data center: traffic request allocation and finding the shortest path. Then the two problems are improved and optimized, and a new load balancing scheme is proposed. Firstly, according to the global network topology information and server load, a wildcard rule AWR is designed to allocate business requests, and then the matching information between the request host and the processing server is obtained. The core idea of this rule is to aggregate requests according to the weight priority of the server to minimize the flow rules. Then, according to the pairing information, an improved algorithm based on the original Dijkstra algorithm, AD, is proposed to select paths and find the shortest path. The key to the improvement of the algorithm is to consider the path bottleneck bandwidth and weight information to make the routing strategy, and then to optimize the bandwidth utilization and delay. In the end, the Fat-Tree topology is built on the virtual experimental platform Mininet, and the experimental data are obtained by testing the different scenarios. Then according to the bandwidth utilization and packet round-trip delay, the superiority of the proposed load balancing scheme is verified.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.0

【相似文献】