Research On Congestion Control Of Heterogeneous Network Based On Network Status Differentiation

With the development of the network technology, the network's data traffic and network applications grow rapidly; even exceed the network's limited resources and processing capacity and thereby the network congestion or collapse occoured. Therefore, how to take effective congestion control measures to keep the network free of congestion becoming a realistic problem that people concern.On the other hand, the new transmission media and network protocol standards are emerging. The network is becoming heterogeneous. Traditional TCP's congestion control measure encountered a series of new challenges. There are two reasons for this problem:first, packet loss is not only caused by network congestion; second, simply classify the network by congestion and non-congestion state is not able to accurately describe the actual state of the network.The accurate differetion of different network status is an effective way to improve the algorithm perfomance. The current prevalent protocols based on network status differetion are TCP Westwood, TCP Vegas and TCP Veno etc. Through the introduncion of new parameters that reflect the network status, calculate and analyze the existing parameters, we get a measurement of the current network quality status. Then we can take more reasonable and targeted congestion control measures.In this paper, reasonable improvements are made on TCP Vegas and TCP Veno based on previous studies. The experiment results show a fine performance. The resarch works are mainly in the following aspects:First, TCP Vegas have the problems that the slow start end prematurely due to the burst growth of the congestion window; Throughput degration occoured in the asymmetric network when the reverse link congested; the uncompatibility problem when coexisting with TCP Reno. Our algorithm differentiates the network status and adjusts the congestion window in both slow start and congestion avoidance dynamically. Experiment results prove that increase the congestion window with a rate between exponential growth and linear growth, and adjust the congestion window by the variation of latency and throughput could effectively improve the transmission performance of TCP Vegas.Second, to solve the problems that when the error rate of the network is high, TCP Veno can not estimate the minimum round-trip delay (BaseRTT) acurately, and TCP Veno can't adapt the burst traffic in network well, which result in a significant performance degration. We proposed TCP Veno+ which differentiates the congestion status of the network by two metrics:delay and throughput and improve TCP Veno's AIMD algorithm. Experiment results prove that the congestion control measurement of the refined algorithm are more suitable for the specific status of network, and have better performance compared to TCP Veno under high error rate environment.Third, TCP Veno has improved traditional TCP protocol greatly in networks contain the wireless connection. But research also indicate that TCP Veno can not use the bandwidth effectively in light load environment, while can not distinguish between congestion packet loss and wireless random packet loss accurately in the heavy load environment and then lead to frequent congestion. We proposed refined TCP Veno-S. It calculates the amount of congestion packet loss and wireless random loss in a certain period. And then use the statistic data to help Veno estimate the network status more accurately. Experiment results prove that the refined algorithm improve the accuracy of packet loss differetion and bandwith untilization, as well as good fairness and stability.The significance of this paper lies in that:Improved algorithms distinguish between different states of the network, and then take corresponding congestion control measures to improve the network performance. It is adapt to the current complex and hetegrous network; On the other hand, in the practical view, the improved algorithm only modify the end system without modifying the routers, and only a slight increase in parameters or use the existing network parameters to distinguish the different states of network, its cost is low; At last, The fairness and compatibility of the improved algorithm also performs well. Overall, The result of this paper has certain practical value.