Intra-protocol Convergence Problem in BBRv2’s Bandwidth Probing

Google introduced Bottleneck Bandwidth Round-trip propagation time (BBR), which is a new concept of congestion control algorithm. BBR measures the available bottleneck bandwidth and minimum round-trip time. It then creates network path models to maximize the delivery rate and minimize latency. However, BBR creates unfairness problem with other TCP flows and excessive packet loss in a small buffer. Google, therefore, updated BBR, creating a new version, called BBRv2, to improve coexistence with other flows and reduce aggressiveness. BBRv2 significantly improved the fairness with loss-based congestion control algorithms and reduced the number of packet retransmissions in a small buffer compared to the existing BBR. However, in this study, we confirmed that multiple BBRv2 flows that enter the same link at different times do not take up bandwidth fairly. Therefore, we evaluate the fairness between identical BBRv2 flows through the Mininet emulation and analyze the reason for this problem. The test results show that when the bottleneck buffer size is smaller than 0.2 BDP, two BBRv2 flows have similar throughput and the number of packet retransmissions greatly reduces compared to BBRv1. However, with a large buffer, the flow starting first occupies more bandwidth.