TCP Congestion Control Based On Accurate Measurement

With the expansion of the Internet,the increase of users,and the sustained growth of user needs and various types of network applications,network data traffic is showing a sharp upward trend,and the network congestion is also become increasely serious.The TCP's congestion control protocol has been the focus of network congestion control research.The current TCP congestion control protocol has evolved in many versions,mainly based on implicit detection or based on display notifications.With the development of various new network structures and different performance requirements under different network environments,the traditional TCP congestion control protocol is difficult to achieve better transmission efficiency.This article has conducted the following researches on the above issues:1)This paper first introduces the basic thoeries of the traditional TCP congestion control protocol and classic algorithms based on implicit detection,such as Reno,Cubic,Westwood,BBR,etc.Through the study,it is found that the packet loss-based congestion control algorithm has sensitivity to random packet loss,and the delay-based algorithm is also difficult to achieve good results in terms of fairness and bandwidth utilization.There are also some methods based on other performance parameters such as the available bandwidth,it is also limited to use the delay variation to obtain the estimated value of this parameter,so the precise granularity is coarser.Based on the above research,this paper aims to obtain more accurate network congestion information through the measurement of available bandwidth in the network scenario with high latency and high packet loss rate,so as to obtain better bandwidth utilization and fair competition.2)Accurate measurement of the available bandwidth is important for the TCP congestion control.In this paper,Pathchirp algorithm is improved based on the feedback of available bandwidth measurements.The Pathchirp-R method is proposed to optimize the rate adjustment of the measurement group.Based on the iRouter equipment and the FAST architecture,an accurate measurement method for delay is developed,which improves the accuracy of delay measurement to the nanosecond level and further improves the accuracy of the available bandwidth measurement.The new method is compared with the Pathchirp method,and the experimental results show that the Pathchirp-R method improves the accuracy and convergence speed of the available bandwidth measurement.3)Based on the above accurate measurement of available bandwidth,this paper improves the Cubic congestion control protocol and proposes the Cubic-M protocol,which mainly includes three parts of optimization: The first is a slow start threshold mechanism based on available bandwidth.By combining the available bandwidth and the slow startthreshold,some sudden increase in the congestion window in the late slow start period is avoided;The second is a packet loss distinguishing mechanism,which is to distinguish whether packet loss is caused by congestion by comparing the sending rate and the available bandwidth;The third is a window sizing mechanism that rapidly increases the utilization of bandwidth when the network is idle.This paper builds an experimental environment that simulates real network performance through TC,to analyzes the bandwidth utilization and fairness of the Cubic-M protocol in different network environments,and to compares it with other TCP congestion control protocols.The experimental results show that the Cubic-M method can maintain high bandwidth utilization in a high-latency and high-drop packet network environment,and the fairness performs well.