Broadband networks are being designed to transport diverse traffic streams with various quality-of-service (QoS) requirements. Accurate characterization of these streams is essential to devising efficient resource allocation strategies that guarantee acceptable QoS without underutilizing network capacity. This project is focused on characterizing compressed video traffic and investigating resource allocation schemes for transporting video traffic over computer networks. New traffic models for VBR video streams will be developed under three strongly related compression algorithms: JPEG, MPEG-1, and MPEG-2. The modeling methodology will be based on the class of M/G/infinity processes, whose versatility will be exploited to capture various forms of traffic correlations. Video sources (single and aggregated) will be characterized at both an entry node and after being multiplexed with other streams (i.e., the departure traffic from a multiplexer). Models for the various layers in hierarchically encoded MPEG-2 video will be developed. The multiplexing performance under video models will be investigated. Corresponding resource allocation strategies for transporting video traffic over broadband networks will be developed. The impact of correlations persistence on the queueing performance under video traffic will be investigated. Several priority scheduling strategies at a multiplexer that is fed by several MPEG-2 input streams will be studied. In connection with the proposed research program, several educational goals will be pursued. New courses on broadband multimedia networks which integrate in their curricula several aspects of the PI's research agenda will be developed. The pedagogical approach in these courses will be a balanced mix of theory and practice and will combine classroom instruction with research-oriented semester projects, hence exposing students (graduate and senior undergraduate) to the intriguing realm of research. Team work will be strongly encouraged in course projects, as it prepares students to future work environments in industry. In an effort to outreach to industry, networking courses taught by the PI will be broadcast over the VideoCampus facility to several industrial sites in the area, allowing many interested engineers and managers to take these courses on their premises. The PI plans to organize a weekly seminar on various topics in networks research, which will feature presentations given by students, faculty, and invited speakers from industry and academia.
