End to end congestion control in high-speed networks
S. Jagannathan · 2003
This paper proposes an adaptive methodology to prevent congestion in packet switched networks such as the Internet, where the internal network nodes convey very little information to the ingress nodes. Two architectures of preventing the congestion are presented: the first one when the traffic arrival rates and bottleneck queue levels are known and the other when these are unknown. In the latter, the network traffic is estimated online using an adaptive system by measuring the buffer occupancy. In both architectures, the congestion is controlled by adjusting the transmission rates of non real-time and certain real-time sources in response to the feedback information so that a desired quality of service (QoS) can be met instead of using the existing additive increase multiplicative decrease (AIMD) algorithm. The QoS is defined in terms of packet loss, transmission delay, network utilization and fairness. Mathematical analysis is given to demonstrate the stability of the closed-loop system. Studies are included to show the effectiveness of the proposed scheme during simulated congestion. The proposed methodology can be readily applied to network planning, designing routing algorithms and transmission links as well as for real-time video and voice data transfer in unicast networks.