BBR-R: Improving BBR performance in multi-flow competition scenarios

Abstract

The development of network infrastructures and the evolving demands of internet services impose higher requirements on congestion control algorithms. Although Google’s BBR algorithm achieves lower latency and higher goodput compared to traditional congestion control algorithms, it still has many issues. BBR sets the congestion window larger than the calculated ideal value to prevent transmission stalling in the presence of delayed and aggregated ACKs. However, in scenarios with multi-flow competition, this compromise on the congestion window leads to large amounts of queued data, causing increased latency and decreased fairness. Additionally, the ProbeRTT mechanism deviates from its original intent. In this study, we analyze the existing issues of the BBR algorithm from a theoretical standpoint and propose the BBR-R algorithm, which incorporates an adaptive sending rate adjustment mechanism and a new ProbeRTT triggering mechanism. While maintaining the ability for dynamic bandwidth exploration, the sending rate is adjusted based on a latency-related factor called Adaptive_RTprop to control the over-injected data. Coupled with the new ProbeRTT triggering mechanism, BBR-R reduces the frequency of entering the ProbeRTT phase and thereby improves transmission stability. In conducted experiments, BBR-R decreases the frequency of entering the ProbeRTT phase in many scenarios, achieves a 41.86% reduction in latency in the dual-flow competition scenario, and improves fairness by 22.79% in the five-flow competition scenario.