Fairness Research And Improvement Of Network Congestion Control Algorithm BBR

The improvement and realization of various network application technologies make people's lives rich and colorful.No matter what powerful application,such as virtual reality or interactive gaming,it finally needs the support of network technology.To ensure the quality of service for users,these applications often require the network to provide high bandwidth,low latency,and fair bandwidth allocation.As a new congestion control algorithm with good performance,BBR was proposed by Google,and it has significantly improved bandwidth utilization and delay compared with commonly used Cubic algorithm.However,with the research on BBR,it was found to have some problems on intra-protocol fairness and inter protocol fairness.In terms of intra-protocol fairness,BBR flows with long round-trip times occupy most of the bandwidth.In terms of inter protocol fairness,when BBR flow coexist with Cubic flow,most of the bandwidth will be occupied by BBR flow.These problems not only lead to unfair bandwidth allocation among BBR streams,but also cause poor compatibility with existing TCP versions.Aiming at the intra-protocol fairness of BBR protocol,this paper proposes AdaptiveBBR.The algorithm has an adaptive bandwidth detection mechanism,which can estimate the amount of backlog data in the current queue by detecting network parameters,and adaptively select the detection gain based on the estimated amount,so as to avoid blindly sending a large amount of extra data in order to detect the available bandwidth.This paper analyzes the theoretical equilibrium point of the scheme by establishing the fluid model.Furthermore,extensive experiments have also verified this scheme has a significant performance improvement over other schemes in terms of fairness and delay in bandwidth allocation.Aiming at the inter protocol fairness of BBR,this paper proposes Friendly-BBR with TCP friendly mechanism.In order to further determine the utilization of network links,Friendly-BBR introduces the gradient of RTT as auxiliary information.When the TCPfriendly mechanism is triggered by packet loss,based on the introduced RTT gradient information,Friendly-BBR will provide feedback on slowing down the transmission rate for packet loss caused by network congestion,which can alleviate the large and frequent packet loss of the network,while improving throughput of the loss-based congestion control.The experimental results show that when Cubic and Friendly-BBR coexist,Cubic's throughput can be improved.Due to the constraints of the TCP friendly mechanism,Friendly-BBR can also greatly decrease the amount of retransmitted data,reduce the frequency and amplitude of RTT jitter.The Adaptive-BBR algorithm makes the allocation of bandwidth between data flows fairer,which guarantees the quality of service for users.Meanwhile,further reduction in latency also makes it more suitable for delay-sensitive application scenarios.The FriendlyBBR algorithm enhances the friendliness of the TCP version and is more compatible with deployed network congestion control algorithms.The reduction in the amount of retransmitted data also improves the effective use of bandwidth.