Filter-based bandwidth estimation for communication networks

Sammanfattning: In recent years, there has been a rapid growth of interest in bandwidth estimation of communication networks. In the context of packet-switched data networks, such as the Internet, the term bandwidth commonly refers to the amount of information a network can deliver per time unit. Knowledge of bandwidth characteristics is of great significance in, e.g., end-to-end admission control, capacity planning, intelligent routing, audio/video streaming, and peer-to-peer applications.For network end users, it is difficult to obtain bandwidth properties of a particular network link or path. Normally, it is only achievable by actively probing the network, and to perform estimation based on received measurements. Probing implies transmission of specific data packets, which are expected to be affected by the characteristics of the utilized network path.In this thesis, the BART (Bandwidth Available in Real-Time) method for real-time estimation of end-to-end available bandwidth is presented and evaluated. The available bandwidth of a network path corresponds to the minimum unused bandwidth capacity among all links between two end hosts. Besides applying active probing, BART uses Kalman filtering for the analysis of received probe packets.BART is investigated both analytically and experimentally. The experiments are carried out in a laboratory network and over the Internet. The opportunity of tuning the Kalman filter with respect to desired tracking ability is studied in more detail, as well as the virtue of enhancing performance by introducing change detection. In general, the results show accurate estimation under various circumstances, even though the method only requires modest computational efforts and minor injections of probe packets.