【作者】 董喜明；

【导师】 余少华；

【作者基本信息】 华中科技大学 ， 计算机应用技术， 2006， 博士

【摘要】 近年来,以太网技术以前所未有的速度向前发展,并向城域网方向迈进。特别是ITU-T X.86标准的提出,首次成功地将以太网和千兆以太网引入到电信传输网中,使以太网技术向城域网方向的发展迈出了关键性的一步。ITU-T X.87的问世,为以太网技术在城域网范围内实现多业务的承载提供了有力的技术保证,推动了城域以太网技术的快速发展。在城域以太网众多的技术中,VPLS(Virtual Private Lan Services,虚拟专用局域网服务)由于技术简单可靠、易于实现而受到人们的青睐。在分析VPLS网络技术特点的基础上,给出了VPLS的网络模型及其转发平台的实现方法。VPLS的网络模型是进行VPLS技术研究的重要理论基础。通过在PE节点之间建立全网状的伪线互连,利用MPLS的标签交换路径建立数据转发隧道,完成对用户业务的承载。分析了VPLS中的MAC地址学习机制,FIB(Forwarding Information Base,转发表)的建立及管理等。分析了具体实现过程中的软/硬件分层结构,并深入探讨了实现VPLS数据转发的状态机。通过大量的试验,对理论研究的结果进行了逐一验证。提出了VPLS网络中的流量工程模型,给出了VPLS可行树的建立算法。VPLS网络中的流量工程是实现VPLS网络优化和保证服务质量的重要方法,有助于实现端到端的QoS控制功能,确保分组在通过VPLS网络时的服务质量。分析了VPLS中为用户提供SLA服务的复杂性,给出了CE节点为实现流量工程采取的QoS预处理,以及流量通过UNI接口进入到服务提供商网络中时PE节点必须完成的主要功能,如对着色帧的处理,UNI接口上的广播帧的转发抑制等。给出了基于Hose模型的带宽分配机制,通过VPLS可行树实现多点到多点的动态带宽分配,以及在链路故障的情况下如何对VPLS可行树中的链路提供路径保护。通过迭代,提出了VPLS中具有时延约束机制的转发算法。点到多点的分组转发机制是VPLS网络中的一个重要研究课题,即:如何真正实现点到多点的转发机制,弥补VPLS技术中由于洪泛带来网络资源的巨大开销。考虑到VPLS将是一种承载实时业务的技术,算法中研究了具有时延约束的点到多点的转发机制,对改善VPLS网络的性能具有重要的理论意义。采用MPLS的显式路由建立点到多点的LSP,减少洪泛,降低PE节点的负担,优化网络的性能。给出了不同情况下转发树的剪枝和加入方法,以满足VPLS成员的动态变化。给出了点到多点MPLS隧道的建立方法,表明设计的方案能够真正地在应用中实现。提出了VPLS中如何通过共享聚合树实现组播技术。VPLS对组播业务的承载是人们关注的一个热点问题。由于VPLS的转发是建立在“分布式”的MPLS交换平台之上,针对两个重要的问题进行了深入探讨:(1)如何获取组播成员信息;(2)如何在PE节点间共享这些成员信息。由于网络规模的差异,分别采用IGMP spooping和PIM Snooping两种方法获取组播成员。给出了组播组和共享聚合树之间的映射算法,以及优化组和树之间由于需要维护转发状态而带来的系统开销。更多还原

【Abstract】 Ethernet technology has been developing much faster than ever before. The ratification of ITU-T X.86 successfully introduces Ethernet into carrier grade telecom network for the first time, and another ITU-T standard, X.87, makes it feasible and reliable for Ethernet to carry multi-services in metropolitan area networks. In recent years, Ethernet has extended itself into metro area networks, becoming the optimal candidate in building metro. Among the many metro-building candidates, VPLS is highly praised for its simplicity, reliability and ease of implementation. It takes the advantages of IP/MPLS backbone to provide customers with an emulated LAN; upon it transparent LAN services can be easily carried out.A VPLS network model and its implementation is first put forward. By building a full-mesh of pseudo-wires between PE nodes and mapping traffic onto MPLS LSP tunnels, VPLS carries different kinds of services for customers. Then it studies the mechanisms for building and managing FIB table, etc. It also describes the architecture for implementing multi-layer VPLS routing and switching platform. Through experiment it concludes that the theoretical model introduced is practical and feasible.A VPLS model for traffic engineering is proposed for the purpose of end-to-end QoS guaranteed packet switching. It points out the complexity when offering customers with SLA services. Some important technical issues, such as the traffic pre-treatment on CE nodes, what steps PE nodes should take when receiving traffic from UNI, including frames coloring, UNI broadcast storm control, etc., are fully addressed. The last issue is the hose model based bandwidth guarantee, in which it illustrates how VPLS feasible tree can be implemented to realize the multipoint-to-multipoint dynamic bandwidth allocation, and how backup path can be computed when link failure occurs in a VPLS feasible tree.A delay-constraint point-to-multipoint forwarding algorithm is proposed to avoid too much flooding in VPLS network. Flooding, if not properly handled, often degrades network performance and wastes resources. With real-time services taken into account, a delay-constraint point-to-multipoint routing mechanism is described through P2MP MPLS LSPs. As complementary, it briefly illustrates how to establish P2MP LSPs,concluding that the proposal can be feasibly implemented.A shared aggregate multicast tree is suggested for multicast services in VPLS. As VPLS is based on“distributed”switching architecture, how to learn multicast membership and how to distribute them among PE nodes are the greatest challenges. It describes both IGMP snooping and PIM snooping to learn multicast membership, and discusses the conflict between multicast states and bandwidth efficiency with shared aggregate multicast trees as an optimized trade-off solution. Through the mapping mechanism between multicast groups and aggregate trees, the overhead for multicast state maintenance can be greatly reduced.更多还原