【作者】 余志洪；

【导师】 雷振明；

【作者基本信息】 北京邮电大学 ， 信号与信息处理， 2007， 博士

【摘要】 对于IP网来说，其最初的设计思想是采用“尽力而为”(Best-Effort)的机制来转发数据分组，对于需要严格服务质量保证的多媒体应用，如果不采取新的技术改善目前的网络环境，则无法实现新业务的大规模应用。目前网络出现的所有服务质量问题，归根结底是由带宽不足带来的。因此，充分增加网络的带宽以使其达到用户需求，曾经是解决因特网服务质量最直接、最有效的办法。然而增加带宽要受到各种因素制约，事实上，新应用的不断出现决定了IP网络带宽的增长几乎无法完全满足业务发展的需求。因此如何合理利用IP网络的有限带宽，使其尽量能够满足那些对QoS要求较为严格的实时应用才是解决问题的关键。目前，电信业务IP化的趋势已经非常明显，特别是NGN和IMS技术的提出和不断成熟。甚至有人提出，下一代电信IP网的核心技术就是IMS。但是，NGN的QoS问题成为每一个标准化组织都不可回避的问题。NGN QoS不同于传统的IP QoS，NGN QoS希望在IP网络上为电信业务提供质量保证。NGN QoS的目标要求QoS解决方案必须将IP的网络处理机制和特定业务的QoS需求联系起来，具体地说，需要解决如下的问题：业务层与网络传输层如何交互和业务的QoS需求如何传递到底层的IP网络上；网络如何满足不同业务特定的QoS需求；网络如何区分业务和非业务流量；如何实现不同业务的QoS隔离；NGN QoS如何在现网上实施等。本论文的主要研究方向是宽带IP网，包括因特网和NGN的QoS的基本理论和有关算法、以及在高速路由器上的具体实现。论文主要研究内容和创新点可概括如下：1)研究了关于QoS的重要基础理论，包括QoS的定义、业务模型的原理及其对网络设计的影响，以及ATM技术的QoS技术特点和功能模块。总结了目前常用的数据源模型，例如语音源的On-Off模型，视频源的MMPP模型和自相似模型等。2)研究了综合服务模型(IntServ)和区分服务模型(DiffServ)两种技术，以及MPLS的基本原理和QoS能力，讨论并比较了各自的优缺点。3)提出了一种适合于Internet使用的端到端QoS模型，它集合了综合服务模型、区分服务模型和MPLS流量工程技术优点。在接入网络一侧采用综合服务技术，而在核心网络采用基于IP的MPLS加上区分服务技术，兼顾了扩展性和QoS保证精度的问题，对具体建网有一定的指导意义。4)提出了一个能够改善TCP连接性能的双向资源预留方案。由于TCP连接的特性，单向的资源预留是不够的。通过仿真，我们可以看到，双向资源预留对TCP连接性能的提高。这也为考虑TCP业务的性能提供了一个思路。5)实现了在核心路由器中，提供满足10Gbps线速处理的多个QoS模块(文中只描述了METER、WRED以及分级报文调度机制)。本核心路由器正在产品化中。在每个模块只有57个时钟周期处理时间的情况下，对微引擎上的每条指令都经过了详细的研究和精确的设计。讨论了在高速网络中OoS机制实现的必要性和难点，本实现方法和机制为高速网的QoS实用以及今后的QoS研究都具有很强的现实与指导意义。6)讨论了各个标准化组织对NGN网络技术的发展研究，分析了业务层与传输层分离的网络结构对IP QoS提出的新问题。7)研究了在NGN网络中，当路由器只提供区分服务的粗粒度QoS保证下，能否对聚合流中的每个流也提供QoS保证。单个业务流和聚合流的丢包率有何不同。根据这些研究，我们总结了相应的设计指导，这对于解决这个NGN QoS问题是一个重要的突破。给出了一种解决方案，基本解决了单个流在聚合流中不能得到很好QoS保证的问题。对NGN中业务层与传输层分离的结构实用化很有帮助。本论文的研究涵盖了ATM、IP、NGN网络的QoS基本理论和相关问题，提出了一些针对性的算法、模型和方案，并在高速路由器中实现了QoS机制。更多还原

【Abstract】 The original idea of IP network is to use the "best effort" mechanism on transmitting packets. So, for the multimedia application which require strict QoS guarantee, if we don’t use new technologies to improve the networking capability, these new applications will not be able to be deployed widely on the Internet.Fundamentally, all of the QoS problems are caused by un-enough bandwidth. The simplest and the most effective way of solving Internet QoS problem is to increase the network bandwidth to meet users’ need. But the incoming new applications always used all of the increasing bandwidth very soon and made the bandwidth will never meet the requirement of the new service deployment. Therefore, how to efficiently use the limited bandwith of IP network is the key to solve the problem and meet the strict QoS requirements of the real-time applications.Right now, the telecommunication network is changing to IP technology, for example the development of NGN and IMS. Someone even said, the key of the next generation telecommunication network is the IMS technology. But the QoS problem of the NGN network is the un-avoidable problem facing every standard and research groups. And NGN introduced new QoS problems, which are different from the normal IP QoS problems, because NGN QoS is expected to provide the telecommunication quality level guarantee.The goal of NGN QoS is to meet the QoS requirements of specific services using IP processing technology. In specific, it needs to solve the following issues: how to coordinate between the service layer and transport layer? How to map the service QoS requirements to the transport IP layer? How to meet the different QoS requirements using the same network? How to separate the impact of different services?, etcIn this dissertation, the fundamental theory and algorithms on network QoS problem are researched and studied. The detailed of QoS mechanisms for high-speed (10Gbps) core router are implemented and discussed. The major research context and the innovative ideas are below:1) Researched on QoS fundmental theories, including the QoS definition, service models and its impact on network design, ATM QoS technologies and the functional model.2) Researched on the IntServ and DiffServ technologies and the basic MPLS ideas and it QoS capabilities. Studied and compared their pros and cons.3) Proposed an end-to-end QoS model for Internet, which embrace the advantages of IntServ, DiffServ and MPLS traffic engineering capabilities. IntServ is used in the access network and DiffServ and MPLS with TE are used in the core network. It has better scalability and different QOS granularities.4) Proposed a new bi-directional resource reservation technology for improving TCP performance. Because of the mechanism of TCP protocol, the one-way resource reservation is not enough. From the simulation, the TCP performance is improved a lot using this technology.5) Implemented several QoS modules with 10Gbps line rate processing capability for core router (discussed METER, WRED and hierarchical packet scheduling mechanism in this dissertation). Every module only has 57-cycle budget so every instruction should be considered carefully. Discussed the difficulities of implementing QoS mechanism, which can provide the reference and guide for future QoS research on broadband network. This implementation is under product development now.6) Discussed the NGN development status of several standardize groups and analyzed the new QoS problems facing the NGN network which has decoupled server layer and transport layer.7) Proposed a new solution to map the service QoS requirement (loss probability) onto DiffServ router. Analyzed the loss probability of a single flow within an aggregated flow and proposed a new mechanism for access control. From the analysis, this solution works well and should be useful for future NGN technology.This dissertation covers lots of QoS theories and mechanisms in ATM, IP and NGN network, proposes several new solutions and mechnisms. The implementation of QoS on high-speed core router is also discussed.更多还原