The Research On Analysis And Application Of Network Traffic Self-Similarity

The popularization of network and outspring of new applications make network traffic increasing rapidly such as VoD and VoIP etc, so that traditional PSTN traffic character will no longer be adopted to current network traffic. The effective method for studying self-similar traffic is to build models which could describe network characters more reality, and be applied to simulation research.In the paper, we introduce research background of self-similar traffic, including several common methods of creating self-similar traffic and estimating Hurst parameters , the cause and influence to network performance of self-similar traffic. Furthermore, having measured local network traffic with intrusion detect system of network center, we deal with packet arrival process and adopt R/S method to estimate Hurst parameter, which is used to distinguish wheather packet arrival process is self-similar or not in statistics. In order to model self-similar network traffic source and make simulation in NS, we use the superposition of an infinite number of source traffic whose durations of the On/Off periods according to Pareto distribution. Comparing with measured traffic based on the same estimating method of Hurst, we validate the effectiveness of the way for creating self-similar traffic. In addition, we make further study on combining traffic source with TCP protocol to understand their effects on self-similarity, by which we can know traffic source self-similarity is independent to network, while flows control algorithm of TCP changing network status automatically could alter the self-similarity of network traffic. However, it cannot lay traffic source self-similarity aside. Meanwhile, the packet drop rate will rise up accompanying with the increasing of UDP loads, and fall down by reducing packet arriving rate, specially at low load. We get similar results by using TCP, but the range of packet dropping rate is smaller than that while using UDP. It shows that TCP with flow control can alleciate packet dropping to some extent and enhance the degree of self-similarity. By increasing buffer length at bottleneck link, packet drop rate will decline somewhat.Last, we discuss new field about network traffic self-similar applied in network security. Considering that the amount of network traffic self-similarityeffects on intrusion detect system, we study the change of self-similarity caused by abnormity of network traffic and build a network traffic model based on user normal behavior, which can alarm on time whether the network is abnormal or not, by means of comparing self-similar traffic's Hurst parameter.