On the Discontinuous System Model for Rate-Based Congestion Control

In this paper we present a cross-cutting design principle for binary decision control (BDC) based congestion control systems, such as AIMD. Due to its discontinuity, constructing the switching function, or the onset condition of the congestion bit, must consider the relative degree and zero dynamics of the system based on sliding mode control (SMC) theory. We showed that the relative degree of the control system of rate-based flow control algorithms is two, so one can drastically reduce the oscillation by using both the queue-length error and its first order time derivative to construct the switching function. When delay is nonnegligible, traffic oscillation increases with the amplitude of increase /decrease in control switching. One can enlarge the attractive zone of the switching manifold and improve the rate of convergence to the switching manifold with larger increments or decrements of sending rate, but this will lead to higher oscillations. We propose a design rule on how to set the amplitude of increase/decrease scheme. Using this rule, we showed quantitatively that one can drastically reduce oscillation of AIMD yet still achieve similar fairness and stability objectives by using smaller decrements and increments of the switching function.