Study Of Metropolitan Area Network Oriented Network Traffic Modeling

Network traffic model acts as the fundamental issue of network performance analysis and network planning. Especially, precise network traffic model is essential to high performance network protocol constructing, network device designing, effective network topology deployment, traffic forecasting, congestion control, load balance, and so on. Due to the surging popularity and increasing utility of the Internet, backbone network is rapidly growing in size, speed and scope, accompanied with the evolving of their traffic properties. As a result, the traditional network traffic model does no longer accordant with traffic of current backbone network. Consequentially, it is a matter of great urgency to research and model the characteristics of backbone network traffic in the near future.In this dissertation, the network traffic of metropolitan area network (MAN) are researched thoroughly on a background of the natural science fund of shannxi province (2005F43) and the project of Xi'An broadband multimedia MAN optimization in 2003 and 2004. The result achieves the expectation and sets up a bridge between the self-similar theory and the practice of network traffic. Main research works and creative contributions in the dissertation include:1. The network traffic is discovered and verified to be accordant with gamma distribution based on analysis of a great deal of traces collected from a real MAN. This result provides a new approach using probability density distribution property for the study of network traffic. Based on it, the influential MWM set up by Rudolf.H Riedi of Rice University (US) is rebuilded to be a gamma distribution and wavelet based model (GWM). Performance of the improved model is showed to be more effective than Riedi's MWM by the experiment, espeically in aspects of statistcal properties of the traffic generated by the model and rationality of the algorithm used in the model.2. Laws appeared in the process of traffic aggregating are exposed and the cause why the MAN network traffic appears gamma distribution is also discovered. These achievements enriche the knowledge of network traffic properties greatly and provide a necessary basis for network deployment with foresight.3. Quantificational relationship between the probability density distribution of the network traffic and the related time scale is discovered, thus, the complex but delicate structure of the MAN traffic is well illustrated. The improvement greatly deepens the understanding on self-similar phenomena and notablely ends the status lasted for more then ten years that the self-similarity of the network traffic can only be described with the simple statistical variables such as the variance and the expectation. As an application instance, it plays important role in the queue anaysis, in the buffer design of network device.4. A gamma distribution and iterative random dichotomy based model (GIRDM) is set up. The experiment shows that the model not only accurately maps the probability density distribution property of the real network traffic accurately, but also has the ability to alter the self-similar degree of the generated network traffic through adjusting the values of the input parameters. Consequentially, performance of the GIRDM is improved greatly than GWM. The model is published at china communication in Aug 2005 and its detailed algorithm will appeared in Journal of System Simulation. Furthermore, its worthiness is emphasized by the fund of Huawei Corp. in 2006, in which, practical projects such as abnormal network traffic detection based on cascade are supported.5. A network traffic sampling tool based on SNMP is elaborately developed to ensure the time scale as precise as ten milliseconds, which is more accurate than before. This advantage makes it possible to catch the details of the traffic porperty in small scale. Traces in the study are collected directly from a real MAN network.