Adaptive controller design for Internet traffic: A control theoretical approach

Abstract

This thesis presents adaptive control techniques for the prevention of end-to-end traffic congestion in the Internet. Instead of using heuristics to solve the congestion problem, we have systematically applied the classical control theory in our controller design in order to always guarantee the stability of the Internet control systems, and instead of using time-domain approaches, we have employed the frequency domain techniques in our control design. To support AQM(Active Queue Management) in the router, we have designed the adaptive P (Proportional) and the adaptive PI (Proportional-Integral) window-based router controllers for the TCP (Transmission Control Protocol) traffic. Our adaptive controllers self-tune only when significant changes in network parameters have caused the phase margin or the gain margin of the AQM control system to drift outside a specified interval. Using the same control theoretical framework, we have provided different controllers to support best-effort streaming media traffic under different network models that we want to study. We first design an adaptive PI rate-based controller in typical network with SISO (Single-Input Single-Output) IP (Internet Protocol) routers. We then employ the DNA (Direct Nyquist Array) method to design an adaptive multi-loop PI rate-based controller under a network with MIMO (Multiple-Input Multiple-Output) IP routers. In addition to window-based and rate-based router control algorithms, we finally design an adaptive utility-based source controller based on pole-placement technique that can provide a relatively smooth source transmission rate.