AT&T Labs Research
About the talk:
Previous attempts to enhance the Internet service model to provide one or more classes of real-time service, embodied in the RSVP and Integrated Services efforts in the IETF, have met with limited success. The need for per-flow signalling messages, per-flow admission control decsisions at network routers, and per-flow state limit the scalability of these proposals and have prevented any signicant deployment. Recently, several researchers have made proposals for endpoint admission control. In endpoint admission control, the hosts probe the network to detect the level of congestion; the host admits the flow only if the detected level of congestion is sufficiently low. By eliminating signalling messages, admission control decisions in the routers, and network state, these proposals attempt to provide a more scalable real-time service.
In this talk, I'll first discuss several architectural issues that guide and constrain the design of endpoint admission control algorithms. Then I'll describe a simulation study that compares the performance of endpoint admission control to traditional router-based admission control, and that examines the tradeoffs between different endpoint admission control designs. The modest performance degradation between the traditional and the endpoint algorithms suggests that a real-time service based on endpoint probing may be viable.
This is joint work with Ed Knightly (Rice), Scott Shenker (ACIRI), Ion Stoica (UCB) and Hui Zhang (CMU)
About the speaker:
Lee Breslau received a PhD in Computer Science from the University of Southern California in 1994. Between 1994 and 1999 he was a member of the Computer Science Lab at Xerox PARC. He joined AT&T in 1999 and is currently the head of the Internetworking Research Department of AT&T Labs--Research in Menlo Park, CA. His research interests include admission control, Internet support for real-time services, web caching and router queue management.
For more information: