Design and Analysis of TCP for the Next Decade Internet

By

Dr. Shao Liu
Postdoctoral Research , Department of Electrical Engineering
Princeton University
 
 

Date: Sept 3, 2008 (Wednesday)

Time: 11:00am - 12:00noon

Venue: Rm. 1009 William MW Mong Engineering Building, CUHK

Abstract :

Standard TCP has worked well for the past two decades, but has showed signs of poor performance for the next decade Internet, as TCP is not good at adapting to the following three trends of Internet evolution: sharply increasing network bandwidth, swarming of real-time streaming traffics, and increasing divergence of services for various applications. I have proposed a new congestion control algorithm called TCP-Illinois for high-speed Internet, and introduced a new Multi-Mode TCP architecture. TCP-Illinois uses packet loss information to determine whether the window size should be increased or decreased, and uses queueing delay information to determine the amount of increment or decrement. It achieves high throughput, allocates the network resource fairly, is incentive compatible with standard TCP, outperforms most other TCP variants, is suitable for large-scale incremental deployment, and has been included into Linux kernel. Multi-Mode TCP architecture is a new TCP design that contains multiple TCP working modes, and a switch engine to switch among the various modes. Different TCP modes are specially designed to better support different service requirements, like real-time streaming, bulk data transfer, etc. In addition to the specialized optimization for particular service types, with this multi-mode design, TCP achieves service differentiation at end users.

Traditional analytical model to study TCP is the fluid model. However, fluid model has several pitfalls, including its incapability to study the synchronization behavior and its incapability to analyze congestion controllers that use both loss and delay. Recently a new stochastic matrix model that can conquer all these difficulties and can model TCP's behavior more accurately that the fluid model. We have extended the stochastic matrix model by replacing a prior impractical assumption with a practical one, derived fairness and stability results for general TCP algorithms, and performed a systematic study of the synchronization behavior.

Biography :

Shao Liu received B.S. degree from Peking University, Beijing, and M.S. and Ph.D. degrees in electrical engineering from University of Illinois at Urbana-Champaign, where his advisors are Prof. R. Srikant and Prof. Tamer Basar. He is currently with Princeton University, where he is a Postdoctoral Researcher in the Department of Electrical Engineering, advised by Prof. Mung Chiang. He is also a visiting researcher in CCS group of Microsoft Research, advised by Dr. Jin Li and Dr. Phil Chou, and he once was a summer Intern in Microsoft Research Cambridge, working with Dr. Milan Vojnovic. His research interests include congestion control for communication networks, peer-to-peer streaming systems, with applications like IPTV, Video on Demand and Video Conferencing, service differentiation and quality of service, rate control for real-time streaming traffic, etc. His recent work includes the design of TCP-Illinois protocol and the analysis on the fundamental performance bounds for peer-to-peer living streaming systems. His email address is shaoliu@princeton.edu, and more information on his recent and past projects is available at http://www.princeton.edu/~shaoliu