No abstract

Our previous work suggests that the current congestion controls for cable networks are inadequate when network participants act strategically. A feedback control could result in greater utilization and fairness of cable networks. In order to design a controller for a network, an observer is needed to evaluate the current network state and traffic flows. This paper proposes an observer model and design for cable networks. The observer design is recast into an online traffic classification problem. Using dynamic network traffic phases observed through simulation, the network classifier is designed with a hidden Markov model to determine the network states.

No abstract