Transmission Performance Optimization Of BBR Unilateral Congestion Control Algorithm

To solve the bandwidth bottleneck of modern network transmission and the buffer expansion problem caused by passive congestion control algorithm,Google proposed the BBR congestion control algorithm in 2016,which is committed to occupying the available bandwidth as much as possible without queuing.The performance of BBR algorithm is superior in network transmission with long delay and large traffic,and its performance is even inferior to that of passive congestion control algorithm in network transmission mainly with short delay and small traffic.In this paper,a series of optimizations are made for the BBR unilateral congestion control algorithm in the wireless local area network of NJUPT.The main contents of the research are as follows:(1)Since the original BBR algorithm cannot adjust the length of the convergence period independently,this paper proposes a method that can determine the length of the convergence period to improve the bandwidth detection sensitivity of the BBR algorithm.Start the additional timer at the beginning of the stabilization phase of bandwidth detection and compare the newly obtained RTT measurement value with the existing minimum RTT value.If the new RTT value is less than the minimum value,set the value of the convergence end indicator to 1.If the algorithm is in the convergence stage at this time,it will jump out of the current cycle immediately,start a new round of speed gain,and restart the timer.If the value of the convergence end indicator is always 0within the specified time,the timer is restarted in the next round of bandwidth detection stabilization phase.The modified algorithm changes the length of the convergence period into a flexible variable with a value between 1 and 6 under the premise of considering fairness.The comparative experimental results show that the modified BBR algorithm improves the sensitivity of bandwidth detection and bandwidth utilization.(2)In order to solve the problem of delay oscillation and empty window caused by not receiving ACK in the STARTUP stage of the BBR congestion control algorithm applied in the campus network,a BBR unilateral adaptation algorithm is proposed.The algorithm only runs on the sender side and is not restricted by network protocols and upper-layer applications.By improving the weighting coefficient of the delay estimator,a delay instantaneous average deviation estimator is designed,and the estimation result is used as the oscillation smoothing factor of the delay estimator,to improve the ability of the delay estimator to deal with severe delay jitter.To compensate for the inevitable empty window problem and sequence number wrapping,a flow state machine and a STARTUP state machine are designed at the sender to maintain a high link throughput.According to the specific transmission situation,the traffic is divided into 6 states: new,blocked,waiting,time-waiting,running,terminated,and according to the traffic feedback,the transmission performance of the STARTUP stage is divided into 3 states: GOOD,NORMAL,BAD.The results show that the improved BBR has better transmission performance than the original BBR algorithm in the STARTUP stage and is better than the passive congestion control algorithm(Reno,CUBIC)which is currently mainly used.(3)The original BBR algorithm cannot cope with the active packet loss of the router,and the transmission performance will drop sharply under high packet loss rate.This paper proposes an improved BBR algorithm,which adds packet loss monitor,packet loss rate threshold and failure stage.The state machine is used to improve the anti-packet loss ability of the algorithm.By creating a memory save set to store the packet loss confirmation information returned by the confirmation packet,once the packet loss rate reaches 1%,it enters the finite fault state machine,and activates the advertisement window attenuation factor,the high outflow identifier,the fault event identifier,and the detection sample.Termination identifier.The fault state machine has 10 states,including the processes of sensing congestion,window attenuation,and congestion recovery.The window attenuation factor beta of the tour period is set to 0.33,and the tour stops detecting new bandwidths.The comparative experimental results show that the optimized algorithm improves the anti-packet loss capability of the BBR algorithm and is better than the CUBIC algorithm under the same packet loss rate.