Research of TCP Based on BP NN and Display Feedback Mechanism

To solve the congestion of network on Internet and improve the throughput, the thesis puts forward an improved algorithm based RED and display feedback mechanism. The average queue length always is close to the max threshold in RED algorithm, and then the delay of the packet is very long. The throughput of the network is low. The new way that using BP NN forecasts the network´s situation in the future and controls the average queue length in the router, can make the average queue length lower and make the network smoothly when congestion happens. Then the packet´s delay be shorter and the throughput be larger. At the same time, the congestion bit can be sent to the source of the network to control the congestion window´s increasing. Then, the amount of the increasing congestion window is dynamic adjusting according to the congestion bit from RED. The result from the simulation in NS2 shows that the improved algorithm has the shorter delay and higher throughput than TCP using RED algorithm.