Modeling the Internet Congestion Control Using a Smith Controller with Input Shaping*

The Internet has shown a great capability of endless growing without incurring congestion collapse. The key of this success lies on its TCP/IP congestion control algorithm. In this paper we use control theoretic analysis to model the Internet flow and congestion control as a time delay system. We show that the self-clocking principle, which is known to be a key component of any stable congestion Internet control algorithm, corresponds to implement a simple proportional controller (P) plus a Smith predictor (SP), which overcomes feedback delays that are due to propagation times. Different variants of TCP congestion control algorithms, such as classic TCP Reno or the recent Westwood TCP, can be modeled in a unified framework by proper input shaping of the P+SP controller structure. Finally, we show that controllers that do not implement the Smith predictor, such as proportional (P) controllers or proportional+ derivative+ integral (PID) controllers, provide an unacceptable sluggish system because they do not implement dead-time compensation.