Understanding of BBRv2: Evaluation and Comparison With BBRv1 Congestion Control Algorithm

Google proposed a bottleneck bandwidth round-trip propagation time (BBR) for TCP to replace the loss-based congestion control algorithms, such as Reno and CUBIC. Unlike the loss-based algorithms, BBR models a network path from source to destination and dynamically controls the sending rate using control parameters, such as pacing rate, congestion window, and quantum to achieve high throughput and low latency. However, many studies have reported performance issues in BBR operation, such as excessive packet loss in shallow buffers, the unfairness among different RTT flows, the unfairness with loss-based algorithms, and so on. Google is developing a new version of BBR, BBRv2, to resolve these performance issues. In this study, we evaluate and compare two versions of BBR on a Mininet emulator and a physical testbed, focusing on whether the BBRv2 alpha can alleviate the performance issues of BBRv1 and whether other novel issues will arise in BBRv2. The experiment results show that BBRv2 improves the unfairness and aggressiveness in small buffer less than 1 BDP. Moreover, multiple BBRv2 flows not only show better fairness in bandwidth sharing, but also reduce the amount of packet retransmissions. However, we observed that the challenging issues such as RTT unfairness, coexistence with loss-based algorithms, and synchronization between BBRv2 flows still exist. This study explores BBRv2’s current behavior in various network scenarios and compares the performance of BBRv2 with the BBRv1 congestion control algorithm.