A prediction-based AQM algorithm for DiffServ networks

DiffServ architecture has recently become a good service model for addressing QoS issues in IT networks because it can implement a set of services on the edge router and has good scalability on the core router. This paper proposes a new active queue management algorithm-PIO based on the prediction of the arrival rate in DiffServ networks. In PIO, the average arrival rate for IN packets and all packets in the next interval are predicted firstly. The NO algorithm decides whether or not to drop the arriving packet according to the queue length for IN packets, the total queue length for all packets, the predicted arrival rate for IN packets and all packets in the next interval. Through the simulation, this paper analyzes the NO performances with respect to the stabilization of the queue, packet loss ratio and attainable throughput under the different subscription ratio and the different number of connections. The simulation results indicate that PIO compared with RIO improves the stabilization of queue and reduces packet loss ratio efficiently whether the network is under-subscribed or over-subscribed, while providing statistical assurance for the bandwidth.