Improving the BBR congestion control algorithm for QUIC

Abstract: Congestion control is an important aspect of network technology, where traffic load is balanced to not cause an overflow in the system. Google has proposed its own protocol, QUIC, which is described as being set to supersede the TCP protocol. QUIC has several advantages, namely having high efficiency and low latency, but also a more flexible congestion control due to it being situated in the user space. To be used in tandem with QUIC, Google developed a new congestion control algorithm called BBR meant to fully exploit these advantages, by reducing latency and increasing throughput. However, as BBR is still a rather new algorithm, there are many different improvements possible to make it more efficient. In this paper, a modified BBR algorithm (mBBR) is proposed, which is comprised of three other algorithms meant to improve BBR by adjusting the otherwise static congestion window and pacing rate gain values based on the round-trip time flow, and is compared to the CUBIC, NewReno, and QUIC/TCP BBR algorithms. mBBR has a greatly lower RTT over CUBIC and NewReno, and reduces it by as much as 20% over the default QUIC BBR algorithm, while maintaining the same level of throughput. This improvement makes mBBR more suitable for usage in RAN-applications and other areas where a lower delay is crucial, without sacrificing network speeds.