Research Of TCP Congestion Control Algorithm Based On Loss Differentiation

With the development of Internet, various wireless communication technologies are becoming matured. Since the wireless network is integrated into the global Internet with its advantages of high flexibility, efficiency and convenience, the Internet appears to have a hybrid characteristic of wired and wireless. However, the traditional TCP protocol is originally designed for the wired networks, when it is applied to the wired/wireless heterogeneous networks, there are not only congestion losses, but also wireless losses due to bit error rate, channel fading and noise interferences of the wireless link. All the lost packets will be regarded as congestion losses and the congestion window will be decreased blindly for the flow control, leading to a deterioration of TCP performance. Therefore, an effective network congestion control mechanism based on loss differentiation between congestion losses and wireless packet losses which has an important research significance becomes the key to optimize the TCP performance.Aiming at solving the problem of TCP performance deterioration in the existing TCP protocol when it is applied in the wired/wireless heterogeneous networks with high bit error rate, the research and algorithm innovations in this paper are mainly analyzed around the end-to-end loss differentiation mechanism, explicit congestion feedback and cross-layer optimization strategy as follows:(1) In order to differentiate the packet losses more accurately, a loss differentiation algorithm based on the dynamic round-trip time(RTT) jitter is proposed. This algorithm adopts a dynamic filtering weight to smooth the RTT jitter samples so as to divide the congestion levels in detail, then differentiates the packet losses and adjusts the congestion window according to the divided congestion grades. Simulation results by NS-2 show that this algorithm provides better performance than the existing TCP variants.(2) In order to overcome the limitations of relying on a single measurement to differentiate packet losses and reduce the influence of the reverse link congestion, a loss differentiation algorithm based on the relative one-way delay with timestamps is discussed. This algorithm is able to access the network congestion degree more comprehensively with a joint of multiple end-to-end measurements, such as the relative one-way delay difference, the number of packet queued in the bottleneck link, and the real-time throughput. The experiment results demonstrate that the traditional algorithm has been improved effectively under the condition of high bit error rate and reverse link congestion.(3) A cross-layer loss differentiation algorithm based on ECN congestion probability and end-to-end measurements is proposed. To improve the accuracy of congestion perception in the end-systems, the improved algorithm firstly counts the number of returned ACKs with ECN feedback marks and predicts the probability of congestion loss in real-time; secondly, by identifying the causes of packet losses through a joint judgment with predicted congestion probability based on ECN and multiple end-to-end measurements, cross-layer between the network layer and the transport layer is optimized. Simulation results show that the improved algorithm achieves higher throughput, better stability, good fairness and friendliness compared to the existing schemes.