Characterization of the busy-hour traffic of IP networks based on their intrinsic features

Internet traffic measurements collected during the busy hour constitute a key tool to evaluate the operation of networks under the heaviest-load case scenarios, and further provide a means to network dimensioning and capacity planning. In this light, this study provides a throughout analysis of the busy-hour traffic measurements of an extensive set of universities, regional networks, and Internet exchange points collected from the Spanish Research and Education Network, RedIRIS. After showing that the traffic volumes observed in the busy hour over time can be modeled by a white Gaussian process, this work takes one step further and examines the influence of the networks’ intrinsic features, mainly population size and access link capacity, on the busy-hour traffic. Well-known statistical methodologies, such as ANOVA and ANCOVA, show that the network size in terms of number of users justifies most of the busy-hour traffic information. We further provide a linear-regression model that adjusts the amount of traffic that each network user contributes to the busy-hour traffic mean values, with a direct application to the problem of link capacity planning of IP networks.