基于OpenFlow的网络负载均衡算法的研究与设计
发布时间：2018-07-14 11:09
【摘要】：随着“云”概念的兴起,云存储、云计算、云视频、云安全等云服务成为了炙手可热的红人。而在这股云浪潮的背后,则是各大云服务提供商建立的大型数据中心,通过成千上万的服务器组成集群,配套对应的存储系统、网络互连设备等各种要素,借助虚拟化的手段充分利用计算、存储和网络等各类资源。在规模如此庞大的数据中心内部每个服务器实例随时都有可能有各种不同应用向网络中注入流量,如何将内部巨大的负载进行均衡使网络资源得到充分利用是个亟待解决的问题。 OpenFlow技术源于斯坦福大学的一个研究性项目,后逐步发展成为SDN的概念,而OpenFlow自然成为了SDN使用最广泛的网络协议。OpenFlow通过集中式的控制器将控制层面从传统交换机中剥离,以下发流表的方式指挥交换机处理网络流量,交换机只负责根据流表转发。OpenFlow这种流表方式的管理使得网络数据的处理层次扁平化,能满足更细粒度的处理要求。 本文针对数据中心常见的胖树网络拓扑,在使用单跳贪婪选路的DLB负载均衡算法基础上,提出了一种基于OpenFlow的改进的GLB负载均衡算法,该算法预先计算出网络中所有主机两两之间存在的所有路径,每条路径包括从源主机至目的主机所需经过的所有链路,算法基于这条路径上所有链路的当前可用带宽为每条路径生成一个权重,用以衡量该路径的均衡程度。当网络中有数据流需要进行选路,算法根据源和目的主机找出所有可用路径,选出权重最高即均衡程度最好的路径作为算法输出,避免了单跳贪婪选路可能获取的局部最优路径。本文使用OpenFlow的方法,在OpenFlow控制器POX平台上以模块的形式实现了这两种算法。然后在经过自定义修改的Mininet仿真平台上,通过搭建胖树拓扑的数据中心网络环境,采用随机流量场景,经过实验得出数据,根据平均带宽利用率、平均报文传输时延和链路负载抖动等性能指标,分析验证了本文所提出的GLB算法在负载均衡性能优于DLB算法。
[Abstract]:With the rise of the concept of "cloud", cloud storage, cloud computing, cloud video, cloud security and other cloud services have become popular. And behind this cloud wave is a large data center set up by major cloud service providers, through thousands of servers forming clusters, matching corresponding storage systems, network interconnection devices, and so on. Use virtualization to make full use of computing, storage and network resources. Within such a large data center, it is possible at any time for each server instance to have a variety of different applications that inject traffic into the network, How to balance the huge internal load and make full use of network resources is an urgent problem to be solved. OpenFlow technology originated from a research project at Stanford University and gradually developed into the concept of SDN. OpenFlow has naturally become the most widely used network protocol in SDN. OpenFlow takes the control level off from the traditional switch through a centralized controller, and directs the switch to process network traffic in the following way. The switch is only responsible for the flow table management based on the flow table forwarding. OpenFlow makes the network data processing level flat and can meet the requirements of finer granularity processing. In this paper, an improved GLB load balancing algorithm based on OpenFlow is proposed based on the single-hop greedy routing DLB load balancing algorithm, which is based on the common fat-tree network topology in the data center. The algorithm calculates in advance all paths that exist between two hosts in the network, and each path includes all the links from the source host to the destination host. Based on the current available bandwidth of all links on this path, the algorithm generates a weight for each path to measure the equilibrium of the path. When there is a data flow in the network to be selected, the algorithm finds out all available paths according to the source and destination host, and selects the path with the highest weight, that is, the best balanced path, as the output of the algorithm. The local optimal path that can be obtained by greedy single hop routing is avoided. In this paper, we use the method of OpenFlow to implement these two algorithms in the form of modules on the POX platform of OpenFlow controller. Then, on the Mininet simulation platform which has been customized and modified, by setting up the data center network environment of the fat tree topology, adopting the random traffic scene, the data is obtained through the experiment, according to the average bandwidth utilization rate, The performance indexes such as average packet transmission delay and link load jitter are analyzed and verified that the proposed GLB algorithm is superior to the DLB algorithm in load balancing performance.
【学位授予单位】：华东师范大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.04
本文编号：2121454
[Abstract]:With the rise of the concept of "cloud", cloud storage, cloud computing, cloud video, cloud security and other cloud services have become popular. And behind this cloud wave is a large data center set up by major cloud service providers, through thousands of servers forming clusters, matching corresponding storage systems, network interconnection devices, and so on. Use virtualization to make full use of computing, storage and network resources. Within such a large data center, it is possible at any time for each server instance to have a variety of different applications that inject traffic into the network, How to balance the huge internal load and make full use of network resources is an urgent problem to be solved. OpenFlow technology originated from a research project at Stanford University and gradually developed into the concept of SDN. OpenFlow has naturally become the most widely used network protocol in SDN. OpenFlow takes the control level off from the traditional switch through a centralized controller, and directs the switch to process network traffic in the following way. The switch is only responsible for the flow table management based on the flow table forwarding. OpenFlow makes the network data processing level flat and can meet the requirements of finer granularity processing. In this paper, an improved GLB load balancing algorithm based on OpenFlow is proposed based on the single-hop greedy routing DLB load balancing algorithm, which is based on the common fat-tree network topology in the data center. The algorithm calculates in advance all paths that exist between two hosts in the network, and each path includes all the links from the source host to the destination host. Based on the current available bandwidth of all links on this path, the algorithm generates a weight for each path to measure the equilibrium of the path. When there is a data flow in the network to be selected, the algorithm finds out all available paths according to the source and destination host, and selects the path with the highest weight, that is, the best balanced path, as the output of the algorithm. The local optimal path that can be obtained by greedy single hop routing is avoided. In this paper, we use the method of OpenFlow to implement these two algorithms in the form of modules on the POX platform of OpenFlow controller. Then, on the Mininet simulation platform which has been customized and modified, by setting up the data center network environment of the fat tree topology, adopting the random traffic scene, the data is obtained through the experiment, according to the average bandwidth utilization rate, The performance indexes such as average packet transmission delay and link load jitter are analyzed and verified that the proposed GLB algorithm is superior to the DLB algorithm in load balancing performance.
【学位授予单位】：华东师范大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.04
【参考文献】
相关期刊论文 前2条
1 张顺淼;邹复民;;软件定义网络研究综述[J];计算机应用研究;2013年08期
2 左青云;陈鸣;赵广松;邢长友;张国敏;蒋培成;;基于OpenFlow的SDN技术研究[J];软件学报;2013年05期
，本文编号：2121454
本文链接：https://www.wllwen.com/guanlilunwen/ydhl/2121454.html
