Modeling Internet Link Delay Based On Measurement

To better understand the internal dynamics and related factors of the Internet, establishing an effective Internet link delay model has great significance. Internet link delay modeling plays a vital role in analyzing and forecasting network performance, designing new protocols, network simulating, performance evaluation, quality assurance, and etc.This thesis described the background and important theoretical and practical sense of Internet link delay modeling, introduced the basic concept and measuring methods of Internet delay, and statistical knowledge involved by delay modeling,on these basis, studied on the distribution model and the potential influencing factors of measured Internet link delay.Based on national measured large-scale network link delay, the thesis used Maximum Likelihood Estimation method to model Internet link delay. Comparing with chosen models, including power-law, Pareto distribution, Weibull distribution, and Lognormal distribution, Gamma distribution is more like the measured data, using Q-Q plot method and Bootstrap method to testify. Then, the thesis defined edgeload and betweenness of Internet links, and described their distributions, studied on the relationship between delay and degree, edgeload, and betweenness, to find that Internet delay is affected by edgeload. Link delay distribution features shows certain trends with edgeload increasing, which shape parameter of Gamma distribution positively related to edgeload, and scale parameter negatively related. Further study showed that the relationship is independent to measuring scope and probing position. Finally, this thesis pictured the curve of delay statistical characteristics in different scope of edgeload.