A new class of Smith predictors for network congestion control

A new class of time-variant Smith predictors for buffer set point control over communication networks is proposed. The new Smith predictor uses two different types of time-variant network delay models, i.e. the forward delay (modeling delay effects on the data stream) and the return delay (modeling delay effects on the buffer occupancy information.) The proposed control scheme tracks the desired buffer level, even under large delay uncertainties, abrupt delay changes and additional disturbances such as unknown buffer depletion rates. The question of stability of the resulting system will also be analyzed. Applications of the proposed scheme range from load balancing in distributed computing to congestion control in WAN networks