BBR-n+ congestion control: Real-time performance with smart exit and advanced AQMs

The Internet is evolving rapidly, and billions of devices are being connected to it at the edge. Performance at the edge matters, and the role of the congestion control mechanism is important. Since the emergence of the Bottleneck Bandwidth and Round-trip propagation time (BBR) algorithm by Google, active research has been going on between BBR and TCP Cubic. BBR-v3 is the third version since its inception, and it has tried to address many of the shortcomings of its earlier versions. The issue of fairness with Cubic and Reno flows, the bandwidth overestimation issue in multi-flow scenarios, high packet transmission rate in shallow buffers, and queuing delays and packet losses during/after its startup phase. Not much research work is yet available on BBR-v3 evaluation with Cubic, especially in a variety of connectivity scenarios, such as wired and wireless together. In this paper, we evaluate BBR-v3 with Cubic and have proposed BBR-n+ (BBR Smart Exit) that refines the generic BBR's (BBR-v3) startup exit by detecting receive window limitations and empirically evaluate its performance with the generic BBR (BBR-v3). The early exit issue of BBR-v3 from the startup phase when the congestion window is receiver-limited is probed, and the Smart Exit algorithm has been proposed. It is the continuation of our work on BBR-n, and we have evaluated it using various performance metrics. The role of modern AQMs has been explored with our testing on Common Applications Kept Enhanced (CAKE) and Flow Queue Controlled Delay (FQ_CoDel) AQMs. Through the experiments, we conclude that BBR-n+ in the receive window limitation test provides a 15-20% median throughput gain over BBR-v3 under different receiver window sizes. A 150 ms reduction in HTTP delay when compared with BBR-v3 and a ~ 300 ms reduction versus Cubic. 17% improvement in ping latency compared to BBR-v3 and 45% with Cubic in the wired scenario with a strenuous load of multiple streams. BBR-n+ outperforms BBR-v3 and Cubic in most of the tested scenarios, though it still exhibits limitations, particularly in achieving fairness when competing against Cubic flows when the number of concurrent streams is eight or more.