软件定义数据中心网络的稳定性分析与主动同步

发布时间：2018-04-21 09:53

本文选题：数据中心网络 + 软件定义网络　；参考：《西南大学》2017年硕士论文

【摘要】：随着云计算网络的快速发展,作为云计算基础框架的数据中心网络吸引了来自学术界和工业界的广泛关注,并得到了高速发展。诸如数据收集、视频内容托管交互、社交网络等网络应用都需要依靠数据中心网络的支撑,在数据中心网络迅猛发展的同时,其各项性能指标所受到的检验也日趋严格。软件定义网络(SDN)技术越来越多地被运用到数据中心网络的管理与构建当中,以更好地实现对数据中心网络的全局部署。在SDN控制中,单个控制器控制的数据中心网络由于控制器本身性能的限制,其可控制的服务器数量相对有限,势必无法满足云计算网络的快速发展对数据中心网络可扩展性的要求。多区域多控制器数据中心网络通过增加控制器数量增强对数据中心网络的整体控制,以满足大型数据中心网络对于可扩展性及稳定性的要求。然而由于控制器只对其所控制的区域网络负责,因而需确保控制器之间信息的一致性才能避免多控制器系统的崩溃,从而需要在控制器之间进行信息同步。现有的控制器之间的同步算法是一种基于时间的周期同步算法,然而不论是采用定时更新还是定时更新与实时更新混合的周期同步算法都可能导致网络信息的不一致性,从而致使整个网络具有较高的丢包率和较差的负载平衡。为了保证网络的一致性,本文提出了一个基于事件触发的主动同步算法,在本文的算法中,控制器之间的信息同步是根据选定的服务器上的负载是否超过其所在区域的所有服务器总负载的平均值而被触发的。因为相较于基于时间的同步而言,基于事件触发同步有更好的灵活性,使得主动同步算法不仅对负载变化更为敏感,也消除了同步与时间之间的关联,避免了整个网络中转发环的出现。仿真实验结果表明,与现有的周期同步算法相比,主动同步算法在丢包率和负载平衡方面有着更好的性能。而在相近的负载平衡性能表现下,主动同步算法相比周期同步算法也有着更少的同步开销。在提出了多个控制器控制下的数据中心网络的主动同步算法之后,本文通过网络中的链路负载作为事件状态,负载的变化视为离散事件,将数据中心网络建立为一个五元离散事件系统,根据李雅普洛夫稳定性分析方法,找出了可以令该系统满足李雅普洛夫渐近稳定的状态函数,证明了SDN控制下的数据中心网络在负载平衡策略下是满足李雅普洛夫渐近稳定的
[Abstract]:With the rapid development of cloud computing network, data center network, as the basic framework of cloud computing, has attracted wide attention from academia and industry, and has been developing rapidly. Network applications such as data collection, video content hosting and interaction, social networks and so on need to rely on the support of data center network. With the rapid development of data center network, the test of its performance index is becoming more and more strict. Software defined Network (SDN) technology has been increasingly applied to the management and construction of data center networks in order to achieve a better global deployment of data center networks. In SDN control, the number of servers that can be controlled by the data center network controlled by a single controller is relatively limited due to the limitation of the performance of the controller itself. It is inevitable that the rapid development of cloud computing network will not meet the requirements of data center network scalability. Multi-area multi-controller data center network enhances the overall control of data center network by increasing the number of controllers to meet the requirements of scalability and stability of large-scale data center network. However, the controller is only responsible for the local network under its control, so it is necessary to ensure the consistency of the information between the controllers to avoid the collapse of the multi-controller system, so it is necessary to synchronize the information between the controllers. The existing synchronization algorithm between controllers is a kind of periodic synchronization algorithm based on time. However, the periodic synchronization algorithm of timing update or timing update and real-time update may lead to the inconsistency of network information. As a result, the whole network has high packet loss rate and poor load balance. In order to ensure the consistency of the network, an active synchronization algorithm based on event triggering is proposed in this paper. Information synchronization between controllers is triggered based on whether the load on the selected server exceeds the average of the total load of all servers in its area. Because event-based synchronization is more flexible than time-based synchronization, the active synchronization algorithm is not only more sensitive to load changes, but also eliminates the correlation between synchronization and time. Avoid the appearance of forwarding ring in the whole network. The simulation results show that the active synchronization algorithm has better performance in packet loss rate and load balancing than the existing periodic synchronization algorithm. Under the similar load balancing performance, the active synchronization algorithm also has less synchronization overhead than the periodic synchronization algorithm. After the active synchronization algorithm of data center network controlled by multiple controllers is proposed, the link load in the network is regarded as the event state, and the variation of the load is regarded as discrete event. The data center network is established as a five-element discrete event system. According to the method of Lyapunov stability analysis, the state function which can satisfy the asymptotic stability of the system is found out. It is proved that the data center network under SDN control is asymptotically stable under load balancing strategy.
【学位授予单位】：西南大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP308;TP393.0

【参考文献】