Network Congestion Control, Active Queue Management And Fairness

With the growth of Internet users and the continuous emergence of new services, Internet has developed into a comprehensive business network which used for transmitting data, voice, video and other multimedia information. But the growth of data traffic causes a great burden to the routing node. As a result, the Internet faces the increasing rate of packet loss and network latency, which result in the serious problem of network congestion.Congestion control is the key to ensure network operation and robust, but because many streams are lack of congestion mechanism in protocol, which occupy more network bandwidth, result in the unfair distribution of resources. Therefore, Active Queue Management (AQM) is hoped to solve the network congestion and guarantee the quality of service. This paper focuses on the stability and fairness of network, and proposes the solution aimed at several AQM defects.(1) Two adaptive AQM scheme, NGL and NRL are proposed based on neuron learning. NGL adopts gradient learning, which is simple and easy to implement. NRL is based on the gradient algorithm, which using reinforcement learning. NRL which has good stability and adaptability can adjust the neuron parameters in accordance with the change of the network environment. The two AQM algorithms do not depend on the object model and have adaptability for nonlinear, time-varying systems.(2) The CHNRL algorithm which based on the combination of NRL algorithm and the flow identification mechanism is proposed to solve the stability problem of queue in CHOKe algorithm. And in order to solve the problem of large packet loss rate in CHOKe algorithm when the load is low, the paper proposes RDT-CHOKe algorithm, which introduce the flow rate as congestion notification and add a threshold to start the flow identification mechanism. The simulations verify the performance of these two algorithms.(3) In the CSFQ algorithm, packet loss strategy is not applicable for TCP flow and fair performance is sensitive to cache, thus R-CSFQ algorithm is proposed. R-CSFQ algorithm which improves the estimate of fair share rateα, combines the loss probability with the occupancy ratio of cache, and even adopts the cache management strategy similar to RED. The simulation results show the R-CSFQ algorithm reduce the over discarding of TCP flow and assure the fairness of queue.