Development of Model Predictive Controller for Congestion Control Problem †


Abstract— Nowadays congestion in computer networks is pointed out as animportant and a challenging problem. TCP (Transmission Control Protocol) has themechanism to avoid congestion in computer networks. TCP detects congestion bychecking acknowledgements or time-out processing and adjusts TCP window sizes ofsenders. However, this control method shows low efficiency in communicationsbecause it is based on a mechanism that avoid congestion after congestion once appearsin computer networks. TCP random early detection RED is another popular congestioncontrol scheme. The fundamental idea behind this control algorithm randomly drops theincoming packets proportional to the average queuing length and to keep the queuinglength to a minimum. To achieve high efficiency and high reliability of communicationsin computer networks, many control strategies based on advanced control theories havebeen introduced to tackle the congestion problem. Model Predictive Control (MPC) isthe only practical control method that takes account of system constraints explicitly, andthe only ‘advanced control’ method to have been adopted widely in industry. MPC is amodel-based method which uses online optimization in real time to determine controlsignals. The solution to optimization problem is usually formulated with the help of aprocess model and measurements. At each control interval, an optimization algorithmattempts to determine the plant dynamics by computing a sequence of control inputvalues satisfying the control specifications. In this work, a planning strategy based onMPC will be developed for congestion control problem. A "preset controllers" approachwill be introduced for such application. The effectiveness of considered controller willassessed in terms of how well it could show good tracking performance, maximizing theutilization of the available bandwidth and to what extent it could cope with systemuncertainties.