Research On Characteristics Of Network Traffic And Design Of A Distributed Passive Measurement System

With the rapid development of network technologies, especially the continuous emergence of new applications based on Internet, the network once acting as an information infrastructure has been becoming the infrastructure of the whole society. Its inhomogeneous, variational development results in the situation that network applications and the network itself are confronted with many challenges and various contradictions. Generally speaking, Internet is just a network usable at present. Although this situation is indeed caused by the initial design spirit of Internet to some extant, the exact reason is that we know little about the characteristics and the essence of the operation of network. This prevents us from using network scientifically and effectively.The Study of the characteristics of network traffic and the passive measurement of network traffic are basic methods and ways to deeply understand the essence of network, and to comprehend the operation of network. They are also the important approaches to improve the performance of network, to optimize the design of network and to implement network engineering. Through network measurement and network simulation, this dissertation studies the characteristics of network traffic, especially traffic on aggregate links. It also analyzes the flow arrival processes, the variance of congestion window of TCP under different service flows and proposes the design of a distributed passive measurement system.The major works of this dissertation as follows:1. The effect of statistical time-division multiplex on the degree of self-similarity is studied. The traffic on backbone is composed of many random incoming flows which are multiplexed by the statistical time-division multiplex technology. Based on comparing the degrees of self-similarity of traffic before and after being aggregated, it can show how much the effect of statistical time-division multiplex on the degree of self-similarity is. The experiment results show that the degree of self-similarity of the aggregated traffic's packet-counts time series is lower than that of any incoming branches traffic. Especially under large network throughput, the decrease of the degree of self-similarity of the aggregated traffic is obvious. For the other factors which would change the degree of self-similarity, the experiments demonstrate that the degree of self-similarity does not decrease as the number of service sources increasing and the size of receival window of TCP has little effect on the change of the degree of self-similarity.