Understanding the Performance of TCP BBRv2 Using FABRIC

This paper presents a performance evaluation of the Bottleneck Bandwidth and Round-trip Time version 2 (BBRv2) TCP congestion control. The experiments are conducted in FABRIC, a national-scale experimental network infrastructure that enables large-scale testing. Google released BBRv2 in 2019 as an improvement over its predecessor, BBRv1. Previous evaluations showed that BBRv2 demonstrates better coexistence with loss-based congestion control algorithms (CCAs), presents low retransmission rates, and produces shorter queueing delays even with large buffers. Evaluations conducted in this paper used FABRIC to reproduce the network conditions observed in Wide Area Networks (WANs). The tests presented in this paper evaluate the throughput as a function of the Round-trip Time (RTT) of BBRv2 compared to various CCAs, the RTT unfairness of BBRv1 and BBRv2, the queue occupancy, and the packet loss rate as a function of the router's buffer size. Additionally, this paper presents and discusses the influence of Active Queue Management (AQM) algorithms to mitigate performance degradation produced by the RTT unfairness and the interaction of different CCAs when sharing a bottleneck link.