Network-wide statistical modeling, prediction,and monitoring of computer traffic

摘要

For fault free functioning of networks it is important to have capacity planning, fault diagnosis, traffic forecasting and provisioning, and efficient routing protocol configuration. Lot of data is available on the network traffic including source, destination, application, size etc. However the data is so voluminous that it cannot be used fully to extract any useful decision parameters. Hence one has to resort to sampling the data to make it hand and manageable. It was found that traffic shows long-range dependence over time, which is closely related to the presence of heavy tails in file sizes, connection durations, and user and application behavior. This study develops a model based on the inexpensive link-level data, but also uses information available in the expensive flow-level data. The model accounts for both temporal and spatial dependencies of traffic, due to the routing mechanism used. The proposed network-wide model is used to examine the network kriging problem, meaning, how one can predict the traffic loads on a set of unobserved links through link-level data obtained from the remaining links in the network. (25 refs.)