TCP congestion control and active queue management mechanisms

Abstract

As the Internet continues to expand in size and diversity, congestion control mechanisms start to represent a more and more important role in meeting the expectations of different applications.Although TCP (Transmission Control Protocol) has its own end-to-end congestion control mechanisms, which act only in a reactive manner, in that congestion control is done after the network is overloaded, IETF (Internet Engineering Task Force) is thinking to deploy additional active queue management mechanisms executed by routers, which can be proactive and drop packets before congestion occurs and thus notify the sources of the incipient congestion.In this study, two different AQM (Active Queue Management) mechanisms and guidelines for their parameter selection are proposed. Both mechanisms are based on fuzzy control theory. The first one is a direct fuzzy controller that performs better than most of the AQM mechanisms proposed in the last few years. The second mechanism is a FMRLC (Fuzzy Model Reference Learning Controller), which adapts to different conditions in the network and shows even better results than the direct one. The simulation scenarios vary from FTP (File Transfer Protocol) connections only to wireless cases and performance metrics investigated for the comparison are the instantaneous queue length, bottleneck link utilization, dropped packets, number of connections and fairness.