Research on Clarifying and Control of the Characteristics of Internet Traffic Based on Monitored Traffic Data

随着网络技术的快速发展和业务的广泛应用，在通信网络中需要支持各种各样的服务质量要求。本课题的目的是基于实际网络监测的流量数据和统计数据，研究网络流量的特征和控制方法，并在广域网和局域网的节点上进行监测，获得了几系列流量数据痕迹。通过对这些监测流量数据的分析，我们证明了两种网络在流量上的长期相关性或所谓的自相似性。此外，我们还发现了一种复数相关性质，而不是精确的自相似性：在短时间尺度上接近于局部Poisson性质，而在长时间尺度上具有强相关性。为了预测具有这种一般相关性的流量的性能，我们提出了一种方法，利用不同的分数布朗运动(FBM)过程来描述不同时间尺度上的相关性，并通过这些FBM过程得到一条复合的性能曲线。在我们的方法中，通过与另一个FBM过程的补偿来考虑相关性突然改变的时间尺度上的性能，以减缓相关性变化的速度。我们还发现，广域骨干网上的流量比局域网上的流量具有更接近于高斯分布的边缘分布。通过对互联网流量特性和服务质量保障方法的深入研究，证明了该方法对于边缘分布接近高斯分布的流量具有更高的精度。

As the fast growing network technology and services for a wide range of network applications, it is necessary to support a large variety of requirements for the quality of service in communication networks. The aim of this project is to study on characteristics and control methods of network traffic, based on traffic data and statistics monitored from real networks.Toward this end, we monitored on nodes of a wide area network and a local area network, obtained several series of traffic data traces. Through our analysis of these monitored traffic data, we evidenced the long-range dependence or so-called the self-similarity in traffic of both kinds of networks. Besides, a complex correlation property rather than the exact self-similarity was found: nearly a local Poisson property in short time scales and a strong correlation in long time scales. In order to predict the performance of traffic with such general correlation properties, we have proposed a method, which uses different fractional Brownian motion (FBM) processes to meat correlation properties in different time scales, and obtains a compound performance curve through these FBM processes. In our method, the performance in time scales with suddenly changed correlation property is considered by the compensation with another FBM process to slower the pace of the change of correlation properties. We have also found that traffic on wide area backbone networks has closer marginal distribution with Gaussian distribution than that on local area networks. And our method using multiple FBM process is proven to be more accurate for traffic with marginal distribution close to Gaussian distribution.The goal of this research has been achieved by the thorough investigation of the traffic characteristics of internet traffic, as well as the approaches for quality of service provisioning.