【作者】 沈永坚；

【导师】 张大方；

【作者基本信息】 湖南大学 ， 计算机软件与理论， 2005， 硕士

【摘要】 Internet已经成为人们生活和经济活动中一个不可或缺的重要组成部分,必须要求有一个可信和可靠的技术来监测和维护网络是否安全、高效、稳定地运行。网络测试和测量技术就是为了满足这一要求而发展起来的测量技术。网络测量技术对及时了解网络实时运行状态和了解网络行为特征十分重要。同时,对重新设计高效的网络系统,重新进行网络性能设施的配置和为不同的网络客户提供优质服务起到了指导性作用。 本文首先针对网络流量行为特征进行分析。在建立网络流量模型时,许多文献都是以网络流量服从泊松分布或马尔科夫链为假设前提,来建立自己的网络流量数学模型。本文将着重对网络流量呈现出的自相似性行为特征进行分析和研究,并根据网络流量服从重尾分布特征,对网络流量行为呈现出自相似性特征的原因进行分析。对ClarkNet-HTTP网络流量集合和Calgary-HTTP网络流量集合的分析结果表明:无论是在工作日期间还是在休息日期间,无论是在上班时间期间还是在下班时间期间,网络流量行为都很好地表现出具有突发性特征的自相似性特征。并且经过进一步的分析发现:网络流量行为表现为自相似性特征与网络在传输文件过程中一次所传输数据文件的大小,用户敲击键盘的时间间隔,突发比特的大小和突发所持续的时间相关,而与网络流量是否处于活跃状态并无直接关系。 根据对网络流量行为特征的分析和对当前一些网络流量数据采样方法研究的分析,同时也为了解决当前许多基于泊松模型的采样算法在对网络流量信息进行采样时,不能很好地解决网络流量具有自相似性特征的突发现象等问题,本文提出了两种网络流量数据采样方法:一种基于历史纪录的网络流量数据采集方法和一种基于集合时间序列的网络流量数据采样方法。它们是通过充分利用当前网络流量的变化特征,即网络流量具有历史记忆性特征和在同一空间时间序列内网络流量变化具有相同的变化特征,而分别提出的。实验结果证明:我们所提出的两种网络流量数据信息采样方法,与传统的网络流量采样方法相比,能够在有效降低网络负荷的情况下,所获得的网络流量样本信息能够更好地反映原始网络流量的变化特征规律。 通过上面几部分内容的分析,我们发现网络流量表现出具有突发性的自相似性特征。本文也将试图从消除网络流量突发现象的角度出发,通过采用流量整形技术和虚拟数据报文段探针来探测网络链路中可用剩余带宽技术,提出一种基于测量的TCP拥塞控制改进算法。使得改进后的算法在一个TCP会话连接过程中,所产生的网络业务流比传统TCP拥塞控制算法所产生的网络业务流显得更加平更多还原

【Abstract】 Internet is becoming the important part of our lives and economy activity. It is necessary that it needs a credible and dependable method and technology to guard and sustain whether the network runs safely and steadily. For this reason, network measurement method and technology was supposed, and it is very important that we find the correct work and behavior of network timely. At the same time, network measurement method and technology takes a proposing and consulting role of planning the network system, redesigning the network performance and providing the different service for different network customers.In this paper, firstly, we will make analysis of network traffic characteristics and behavior. When modeling the network traffic characteristics, many papers make the model of the network traffic on the basis that distribution of network traffic is Poisson distribution or of the Markov Chains. We will first emphasize on the analysis of the network traffic here. We also know that network traffic acts as self-similarly characteristics since the distribution of the traffic is a heavy-tailed distribution. Here we will also analyze the reason that the network traffic acts as self-similarly characteristics. Through the result of analysis on the ClarkNet-HTTP traffic aggregate and Calgary-HTTP traffic aggregate, it shows that the network traffic characteristics acts as self-similarly and self-similarly characteristics depends on file size, the time interval of knocking the keyboard, burst size, and burst time length, not depends on whether network traffic is on active state.Secondly, according to the result of the network traffic characteristics analysis, we will consult some current methods of network traffic sample and suppose two new network traffic sample methods: a network flow data sampling method based on history memory and a measurement and analysis of network flow characterization based on aggregated Time-serial. Through the change characteristics of current network traffic, they are both supposed on based of the traffic change characteristics with history memory and the same change characteristics within same space-time serial aggregate. According to result of the experiment, the two network traffic sampling methods supposed in this paper can do better in comparison to the current methods of network traffic sample on performance and reduce the network traffic sample amount.更多还原