Research Of Congestion Control Algorithms On IP Network

With the evolvement of Internet based on TCP/IP, the scale, users and traffics of it have experienced an explosive growth since 1990's. The network congestion has become more serious and complex due to the ever-increasing network application types and dynamic network parameters such as active sessions and round trip time. Congestion often results in decline of quality of service (QoS) in terms of transmission delay and throughput, while the network resource utilization like bandwidth and buffers are also affected seriously. The congestion control is always a hot spot in the field of network research.Internet primarily relies on TCP end-to-end congestion control to avoid network congestion. But TCP strategies couldn't meet many various demands of every application on the complex network. In fact, it will be more effective for detecting and preventing congestion if the routers perform queue management schemes. Congestion control strategies for the routers have been discussed widely. Multimedia applications such as video and audio have come forth continuously in recent years. UDP protocol is employed to transfer multimedia flows. However, UDP flows are not friendly to TCP flows because they have no congestion control scheme, so it's necessary to develop congestion control schemes for UDP flows. These schemes must not only support certain QoS but also be TCP-friendly.Congestion control strategies for IP based networks are introduced in detail from two different layers of routers and end host. Systematic research to meet practical requirements of IP based networks is performed in depth. The primary works and innovations of this paper include but not limited to:(1) Considering the nonlinear, time-varying and uncertain characteristics of networks, a new active queue management (AQM) algorithm, i.e. NNPC-AQM was proposed based on predictive control theory, which requires less model accuracy. In order to realize fast control, a predictor was constructed using two-layer linear neural network to predict the future queue length, and a controller was composed using two layers of nonlinear neural network to optimize the next control volume, i.e. drop probability. By exploiting feedback adjustment, the proposed algorithm requires less model accuracy, together with decreased computation and fastened speed. Simulation results show that this algorithm possesses robustness against variances of network parameters.(2) With the increasing applications of real-time multimedia on Internet, an ECN-based TCP-friendly Congestion Control (ETCC) scheme was proposed. Thescheme emphasizes on the improvement of end host with the support of extended router. RED was adopted in the router, and early congestion state was notified to the sender using ECN. Sending rate could be adjusted according to the rate scheme in the end host. Simulations showed that ETCC can adjust sending rate smoothly, decrease delay and packet loss ratio while maintaining good TCP-friendliness.Congestion control is a sophisticated task and this paper does research on some of it. The only way to resolve congestion completely is to utilize various strategies to control the congestion from multiple aspects of the whole network.