Dynamics of Internet Transport Protocols

Friday, March 25, 2005

Dr. Nageswara S. V. Rao

Center for Engineering Science Advanced Research
Computer Science and Mathematics Division
Oak Ridge National Laboratory
Oak Ridge, TN  37831-6016
raons@ornl.gov

Abstract:

A number of next generation Internet applications such as computational
steering, interactive visualization, and instrument control, require protocols
with stable dynamics. The Additive Increase and Multiplicative Decrease (AIMD)
congestion control method of the widely deployed TCP is analyzed. By utilizing
analytical methods, TCP dynamics are shown to contain chaotic dynamics. Such
dynamics are illustrated in simulations under very simple background traffic
conditions such as a steady UDP flow. In actual Internet traces, both chaotic
and stochastic components are shown to be integral components of TCP dynamics.
The former is due to the high non-linearity of AIMD congestion control and the
latter is due to TCP's response to the inherent stochastic nature of Internet
traffic. 

Two new window-based transport protocols are proposed based on the stochastic
approximation method, which are analytically shown to stabilize at specified
throughput levels. They avoid the chaotic regions by not utilizing the AIMD
method, and they explicitly account for the stochastic nature of the traffic by
throttling the adjustment step sizes as per the classical Robbins-Monro
conditions. In several types of network connections these protocols achieved
robust dynamics in presence of different types of competing traffic.
Additionally, DOE UltraScience Net and NSF CHEETAH projects, and a
probabilistic quickest path algorithm are briefly described.

Biography:

Nageswara S.V. Rao received the B. Tech in electronics and communications
engineering from Regional Engineering College, Warangal, India, in 1982,
the ME from the School of Automation, Indian Institute of Science,
Bangalore, in 1984, and the PhD in computer science from Louisiana
State University in 1988. He is currently a distinguished research staff
member at Oak Ridge National Laboratory where he joined in 1993. He was an
assistant professor of computer science at Old Dominion University during
1988-1993. His research interests include network transport dynamics,
statistical approaches to transport control, information and sensor
fusion, robot navigation, and fault diagnosis. He is a senior member
of IEEE.