An Analysis of the Queueing Delays and Throughput of the TCP BBR Congestion Control in NS-3

Abstract: BBR is a congestion control recently proposed by Google, unlike the traditional congestion control which uses packet loss as the signal of congestion, BBR uses the estimation of bottleneck bandwidth to control the sending rate. However, recent work shows that BBR suffers from a variety of problems such as large queuing delays with multiple flows in the network. Most of the existing work in this area has so far focused on the performance analysis of BBR. Despite these efforts, there exists still a lack of understanding on how to improve the performance of BBR in different scenarios. In this paper, we first present the behaviour of the original BBR in the Network Simulator 3 (ns-3), then, we provide an improvement that carefully adjusts the pacing rate based on the RTT of the flow, finally, to validate our method, we run simulations varying different bottlenecks, latency, and numbers of flows in both small and large buffer size scenarios on ns-3 network simulator. The results show that our improvement can significantly reduce the queuing delay on the bottleneck at a very small cost of throughput in large buffer scenarios, and also achieve less than 0.1% retransmission rate in small buffer scenarios.