Throughput Distribution and Stabilization on TCP BBR Connections

Abstract

Abstract—TCP BBR (Bottleneck Bandwidth and Round-trip time) is one of the most promising transport layer algorithms in the near future. This algorithm provides higher throughput than existing algorithms. However, it was reported that the throughput fairness between TCP BBR connections that share the bottleneck link is not high in some cases. In this situation, the throughputs of some TCP BBR connections are not high enough. We think that this should be solved for TCP BBR to become popular. In this paper, we focus on the TCP BBR implementation of the Linux kernel and discuss the throughput fairness between TCP BBR connections. First, we evaluate the throughput fairness and show that the fairness is not high in some cases. Second, we reveal that the cause of this unfairness. Third, we propose a method for improving fairness by fixing a parameter in this implementation. Fourth, we evaluate the proposed method and show that the method can improve fairness significantly. In the cases of optimized parameters, the fairness index improved more than four times.