Toronto Networking Seminar

Congestion Responsiveness of Internet Traffic:
A fresh look at an old problem

Constantine Dovrolis
College of Computing
Georgia Institute of Technology

Date:  November 18,  3pm
Location: BA1210 (Bahen Center)


The stability of Internet traffic has been attributed to TCP's congestion control. In this paper, we argue that congestion control for individual transfers is not sufficient to produce stable aggregate traffic. The offered load at a network link is generated from users/applications that generate finite-length transfers or groups of transfers (``sessions''). A traffic aggregate is congestion responsive if the session arrival rate decreases when the offered load exceeds the link capacity. We examine two traffic generation models at the session layer. First, a closed-loop model where each user from a fixed-size population can generate a new session only after the completion of her previous session. Second, an open-loop model where sessions arrive independently of previous sessions. These two models produce traffic with very different congestion responsiveness, even if each connection is controlled by TCP. We introduce two metrics to quantify the congestion responsiveness of a traffic aggregate: elasticity and instability coefficient. The elasticity measures the degree by which a traffic aggregate backs off upon a congestion event. The instability coefficient measures the rate with which the number of active sessions increases during a congestion event. Overall, the congestion responsiveness of an aggregate depends on the fraction of traffic that can be modeled as closed-loop (CTR). We describe a procedure to estimate the CTR at an Internet link from packet traces, at least for a significant fraction of the TCP load. Our measurements at several Internet links show that the CTR is significantly high (70-90%). This implies that a major reason for the stability of the Internet today is the fact that most traffic is responsive to congestion at the session generation layer. Finally, we discuss some implications of our observations in several areas of networking research and practice.


Constantine Dovrolis is an Assistant Professor at the College of Computing of the Georgia Institute of Technology. He received the Computer Engineering degree from the Technical University of Crete (Greece) in 1995, the M.S. degree from the University of Rochester in 1996, and the Ph.D. degree from the University of Wisconsin-Madison in 2000. His research interests include methodologies and applications of network measurements, bandwidth estimation algorithms and tools, overlay networks, service differentiation, and network problem diagnosis. He received the NSF CAREER award in 2004.