Cascaded Fuzzy Congestion Controller for TCP/IP Traffic

Abstract

This paper presents a fuzzy congestion control solution for active queue management in wireline packet switched networks, that use TCP to solve end-to-end flow control. Queue management is a trade-off between link utilisation, delay and the loss of packets. The targets of it are to reduce the average queue length in routers, decrease the end-to-end delay and reduce the packet loss probability by preventing buffer overflows. In the presented solution, a fuzzy controller regulates the rejection probability of received packets in routers to prevent buffer overflow and the rejection of all packets during congestion. Using simulations a fuzzy controller is compared with the traditional random early discard (RED) algorithm. Pareto distributed traffic was used in the comparison for the evaluation of controllers. The results showed that the fuzzy controller decreased the packet loss rate with lower buffer occupation and increased the link utilisation and throughput. It also decreased control traffic between the transmitting nodes and routers via decreasing the number of ICMP (Internet Control Message Protocol) messages. In addition, by keeping the average queue size small, the developed algorithm had the ability to accommodate itself more effectively to very bursty traffic than the RED algorithm without dropping packets.