【作者】 朱畅华；

【导师】 李建东； 裴昌幸；

【作者基本信息】 西安电子科技大学 ， 通信与信息系统， 2004， 博士

【摘要】 随着IP网络规模和用户数目的不断扩大，业务种类日趋多样化，加之对服务质量、网络与信息安全的要求越来越高，给其运营、维护和管理提出更高的要求。同时，掌握IP网络的运行规律，对网络设备和协议的研究开发具有重要意义，为此促使了网络测量和业务性能的研究。 本文围绕IP网络测量和业务性能两个方面展开研究，力图构造有效的网络测量基础结构，获得IP网络的性能指标，掌握IP网络的业务性能，理解其运行规律，建立其行为模型，以进一步提高其服务质量，改善安全性，实现高效的运营维护，并为网络新协议的开发奠定基础。本文的主要内容和结果如下： 1．研究了Internet单向时延的测量方法及其动力学特征。提出了一种基于非线性规划的方法估计收发时钟的频差和相对偏差，克服了常用的端到端时延测量方法大多依赖于GPS接收机或采用NTP协议来实现收发端时钟的同步，而GPS接收机价格较高且与接收环境有关、NTP协议的精度不能满足要求的缺点。我们在几条不同的链路上进行了测试，结果表明该方法能有效消除收发时钟不同步对时延测量的影响。此外，通过分析多条链路的测试结果，发现Internet端到端时延具有非线性、非平稳的特征，而且不同链路其自相似性(Self-Similarity)程度不同。通过构造合适的网络结构及泛函神经元函数，利用泛函网络进行系统辨识，准确描述了端到端时延的非线性特征，同时对泛函神经元函数的参数进行自适应修正，适应了时延的非平稳性。对两条不同种类链路实测的端到端时延数据进行验证，结果表明所构造的网络结构以及提出的自适应算法能准确描述Internet端到端时延的动力学特征。 2．研究了端到端瓶颈带宽和可用带宽的测量方法。在分析几种主要的网络瓶颈带宽测量方法及其存在的问题的基础上，提出了一种基于多分组(Multi-Packet)技术、考查分组到达间隔特征(平稳性，标准差与均值之比)的新方法，仿真试验结果表明该方法能有效地克服分组对(Packet Pair)技术无法判断时间压缩和时间扩展造成的误差，相比分组链技术精度得以提高，并且适合于网络重负载的情况。此外，基于自感应拥塞的思想提出一种新的端到端可用带宽测量方法，该方法采用步进式增加探测分组发送速率的方案，还定义了一个判决标准：归一化加权时延增量(NWID)。采用滑动判决方法，一次测量便可得到可用带宽值。仿真试验结果表明：与pathChirp方法相比，适量增加了测量开销，但其测量精度明显提高。 3．研究了802.11WLAN接入点汇聚的业务量特征和WLAN QoS问题。经过仿真实验发现：采用IEEE 802.11协议的无线局域网，其无线接入点(AP)在饱和情况下汇聚的业务量数据具有多分形/多尺度(mul ti一fractal/multi一SCale)特征。利用多分形业务量排队分析的最新结果，研究了无线接入点的业务量丢弃率与缓冲区大小的关系，并进行了仿真验证。此外，针对于少数慢速终端导致整个WLAN性能下降以及兼顾端到端Qos的情况，提出了一种接入控制方法—面向QoS的混合接入控制(QoS一oriented Hybrid Admission Control，QHAC)策略，通过限制慢速终端的分组请求间隔来降低对其它移动台的性能影响。此外，还测量端到端路径的性能，与需要接入的新移动台的服务质量需求进行对比来判决新移动台是否获得竞争权限。仿真结果表明该算法能有效提高工EEE 802.1 1 WLAN的性能。 4.研究了GPRS承载IP业务的性能。在分析GPRS RLC/MAC工作过程的基础上，基于D一BMAP/D/l排队模型分析GPRS数据业务在简化的一步接入模式下RLC/MAC缓冲区队列的平均长度、小分组(Mini一Packet)的平均时延和丢弃率等性能。考虑了不同服务速率、定时器大小和缓冲区容量时的情况。分析结果表明:当定时器TJl:2时间变长时，时延缩短，丢弃率降低;处于空闲状态的移动台产生新会话 (Session)的概率越大，时延越长;缓冲区容量增加，丢弃率明显降低。然而，随着每个Session中分组呼叫数目的变化，时延和丢弃率的变化不显著。本文的特点在于采用了矩母函数和概率母函数理论，运算量大为降低。 5，提出了一种分布式网络测量与分析基础架构(Distributed NetworkMeasurement and AnalysiS Infrastrueture，DNMAI)，采用层次化模型，模块化设计、易扩展。与现有测量框架相比，DNMA工强调了根据用户需求定制测量与分析方案的能力，完善了分析方法和应用场景，可灵活组织单点或分布式多点测量。文中给出其框架模型、功能部件、实现的关键技术和解决方案，并以端到端时延测量与分析需求为例说明了DNMAI的应用。此外，探讨了网络测量在无线网络入侵检测和脆弱性分析方面的应用，给出了一个分布式无线入侵检测和脆弱性分析系统(W工D脚As)模型，作为DN琳工的一个特例。关键词: 网络测量，端到端时延测量，端到端带宽估计，业务性能分析，分布式结构更多还原

【Abstract】 With the explosive development of the scale, the number of users, and services types of IP networks, more rigorous schemes to operation, administration and maintenance are required for building network infrastructure with high QOS and perfect information security. And it is important to master the behavior of IP network for research and development of network device and protocols. Thus, the research on network measurement and traffic performance is paid more attention.This paper focused on IP network measurements and traffic performance analysis, which aims at building an effective network measurement infrastructure, and mastering the end-to-end and traffic performance characteristics of IP networks. The final objects are to give better comprehension of network behaviors, to improve the QoS, to solid the security ability and to build a basis for development of new protocols. The main work and results of this paper are as follows:1. The conventional methods to measure one-way delay rely on GPS receiver or NTP protocol to keep synchronization between sender’s clock and receiver’s clock. However, GPS receiver is not ubiquitous in price and environment and NTP protocol is low in accuracy. A new method based on nonlinear programming, which can remove the skew and offset between the two clocks, is presented. Measurement results show that the method works well. Nonstationarity and nonlinearity are found by analysis of various delay series measured from different links. And different types of links, different Self-Similarity. By constructing appropriate network architecture and neural functions, functional networks can be used to identify Internet end-to-end nonlinear delay time series. Furthermore, by using adaptive parameter studying algorithm, the nonstationarity can also be well modeled. The results show that the functional network architecture can precisely characterize the Internet end-to-end delay dynamics.2. Several measurement methods of network bottleneck bandwidth, with their advantages and disadvantages are discussed. An enhanced method is presented in which new decision rules are added into multi-packet method by analyzing the stationarity and the ratio of standard deviation to mean of probing packets interarrival time series. The simulation results show that the method eliminates the error caused by time compressing or expanding which packet pair method can’t overcome, improves the precision than multi-packet method and works well under heavy loads. A new active measurement approach to Internet end-to-end available bandwidth estimation is proposed, which is based on the concept of self-induced congestion. In this approach,the transmission rates of probing packets are increased step by step and a new decision criteria, NWID (Normalized Weighted Increased Delay), is defined. Another idea is that sliding-window decision is adopted. Simulation results show that the measurement precision is improved with slightly adding probing overhead compared with pathchirp.3. Simulation results show that the aggregate traffic at Access Point (AP) of IEEE 802.11 WLAN can be better classified as a multifractal or multiscale stochastic process. Based on the new results on multiscale queuing analysis, the performance of the traffic loss rate with variable buffer size is analyzed. The simulation results are consistent with the numerical results. Furthermore, a new QoS-oriented Hybrid Admission Control (QHAC) scheme in IEEE 802.11 WLAN is proposed, in which the influence of the mobile station with slow speed on others is decreased by limiting its inter-request time. In addition, end-to-end performance metrics are also measured and applied to make a decision for admission control of new request. Simulation results show that QHAC scheme can efficiently improve the performance of IEEE 802.11 WLAN.4. A D-BMAP/D/1 queuing model is applied to analyzing the performance of GPRS data services, such as mean RLC/MAC queue length, mean delay and loss rate of mini-packets, in which a simplified one-phase access procedure, different ser更多还原