New Traffic Models for Internet Connections and VBR Video Traffic This project examines new models for application and connec- tion level traffic on the Internet. The objective is to better understand the impact of this traffic on network congestion. Features of Internet traffic such as its non- Poisson statistics and long range dependence are a point of departure from typical traffic on circuit switched networks. As packet traffic progresses from the user application to its destination across a network, it undergoes a variety of transformations. These linear and/or nonlinear transforma- tions give rise to structural features in an otherwise ran- dom traffic stream. The result is high variability in the network traffic mix. Identifying and controlling the source of these traffic features is important in real-time resource allocation and network traffic engineering. The work in this project uses measured network data to identify the feature space required to model Internet usage. Traffic generation from end users, network protocols and video-coders are con- sidered. Non-linear time-series models will be used to characterize connection and application level traffic. These models have the flexibility to address a range of traffic types from short-range dependent stochastic processes to those with self-similar and deterministic features. At the application level, particular attention is paid to the gen- eration of layered variable bit rate video and into the in- fluence of encoding parameters on traffic characteristics, model parameters and performance. The time-series modeling framework will be used to design optimal rate control algo- rithms with perceptual and network driven cost constraints. The traffic models obtained will allow one to assess and control end systems and provide network level control for problematic traffic sources. The results will also provide quantitative evaluation of what types of traffic can be su- perposed to yield suitable statistical multiplexing gains. In conjunction with the research effort, a graduate and un- dergraduate curriculum in the network performance area will be developed. The courses and projects will support the development of end-to-end problem solving skills. The aim is to integrate research and interdisciplinary ideas from engineering and physical sciences into the solution of telecommunications related problems. Ongoing activities under this project can be found at http://morse.uml.edu/-kchandra/career.
