Hierarchical Model Predictive Traffic Engineering

Hierarchical traffic control is a promising approach for improving scalability in the face of network size. In this scheme, multiple controllers are introduced in a network, and these hierarchically decide operations. At the bottom layer, controllers decide specific operations in a small area, while the controllers at the upper layer decide inter-area operations using abstracted information from the lower layers. These controllers depend mutually on controllers in other layers, which may cause control oscillations, disturbing the appropriate network state. The common way to handle such oscillations is to set the control interval of the upper layer to a large value. This approach, however, causes another problem: the delay of upper level operations relative to environmental changes. To solve this problem, we introduce the concept of model predictive control (MPC) to hierarchical network control. In this method, each controller gradually changes operations based on the predicted future network state. By predicting the behavior of other controllers in the upper/lower layers, the controller can smoothly shift to the suitable operations. Furthermore, the impact of a prediction error can be reduced by avoiding significant changes in operations. In this paper, we develop MPC-based hierarchical network control for effective hierarchical traffic engineering (TE) . Through extensive simulation, we show that the MPC-based hierarchical TE can avoid congestion even in the cases where the existing TE method of setting long control intervals for the upper layer cannot accommodate dynamically changing traffic owing to operational delay.