Title: Modeling Access Patterns for Evaluation and Enhancement of Streaming Delivery Mechanisms

Abstract: The Internet is a promising and convenient channel for the delivery of streaming media (such as audio/video). Modeling user access patterns and generating synthetic workloads are fundamentally important to the performance evaluation and enhancement of streaming delivery mechanisms. In this talk, I will first describe my work on modeling streaming access patterns and the GISMO tool for generating synthetic workloads. Streaming access exhibits various characteristics such as Zipf-like popularity distribution, temporal locality, and user inter- activities. GISMO captures them and enables realistic evaluation of streaming techniques in a scalable way. Then, I will demonstrate why modeling access patterns is critical to the evaluation and enhancement of streaming delivery mechanisms. An example is multicast-based streaming where the server utilizes network multicast capability to reduce its bandwidth requirement. My analysis indicates that to provide immediate service to users, server bandwidth requirement grows as the square root of user concurrency level if access is interactive, much higher than the logarithmic bandwidth requirement if access is sequential.

Speaker Bio Sketch: Shudong Jin is a research fellow at Boston University. He received a Ph.D. degree in Computer Science from Boston University, in 2003. His main research interests include Web and streaming media traffic characterization, Internet modeling, and performance evaluation of large-scale content delivery techniques. He is also interested in network transport protocols, Web caching, and storage management, data and knowledge management. Before joining Boston University, Shudong Jin received Bachelor's and Master's degrees in Computer Science from Huanzhong University of Science and Technology, Wuhan, China, and worked in a database and multimedia research lab there. During the years of his Ph.D study, he worked at IBM T.J. Watson Research Center in 2000 and 2001, and received the IBM Ph.D. research fellowship in year 2001-2002.