Measurement-Based Analysis And Modeling Of Internet Delay

With the rapid development of network technology and wider applications of Internet services, the Internet is getting increasingly bulky. The supplies of the Internet and requirements of customers hence are becoming more and more diverse and complex, so there are more requirements of network performance and Quality of Service being brought forward. The delay is one of the most important parameters in evaluating the performance of networks. Moreover, understanding of delay can also significantly help promote the QoS.Based on the measured delay data, the information of network topology, network status and some related factors which may impact the delay, this paper explores the relationship between the delay and the factors, and generalizes it. Moreover, it proposes an effective end-to-end delay estimating approach for modeling in order to predict the delay in Internet.Through a long-term periodic measurement on routing information of hundreds of thousands IP addresses, we have obtained a countrywide router-level topology map and the delay data corresponding to this topology. We use these data and the ISP information released by APNIC to research on delay. With statistical methodology, such as curve fitting, we study the probability distribution of the node degree, delay and hops. Further more, we analyze the impact of hops, ASes and common prefix of IPs on delay. And we try to explore the relationships between betweenness centrality, load, degree and delay with data mining methods. We found that betweenness and load restrict the distribution range of delay.Finally, this paper represents two delay models. End-to-end delay predicting model can predict delay between arbitrary IPs in short-term through less measurements, even without measuring. Inspired by GNP (Global Network Position) model, we build another model called Internet delay synthesis model which treats delay as distance between points and then form a delay space. As a result, acquiring of delay becomes a calculation of the distance between points. With the theory of random point process, we can work out the delay data on a larger scale, comparing to using the input measurement delay data.