Understanding the Modern Internet's Heterogeneous Congestion Control Landscape
Abstract:
Research in Internet congestion control has seen a renaissance in the past few years driven by two key developments. In 2016, Google proposed and deployed BBR, a congestion control algorithm that represents a departure from traditional loss-based algorithms like CUBIC and New Reno. Internet transport is also moving to the userspace, with the adoption of QUIC, a new transport stack that is already widely deployed and is set to be the default with HTTP3. While both these developments pose their own unique challenges, they both also introduce a large amount of heterogeneity in the Internet's congestion control landscape.
One of the reasons Internet congestion control has largely remained stable in the past is that it has benefited from a homogeneous ecosystem of well-understood AIMD congestion control algorithms. However, the recent advancements in BBR and QUIC will challenge this paradigm. In this seminar, we examine the impact these two developments will have on the Internet's congestion control landscape, and how they will likely influence its evolution in the near future. We present a 5-year study into how congestion control on the Internet is evolving in response to the deployment of BBR and QUIC from 2019 to 2023, with snapshots of the composition of congestion control algorithms deployed on the Internet in 2019 and late 2023. Our results show that BBR is a dominant force on the Internet and is generally preferred by websites for the delivery of video and other dynamic web traffic. While the general consensus is that BBR can provide superior throughput on the Internet today, our mathematical models suggest that this performance advantage over traditional CCAs like CUBIC is going to wane as more people deploy BBR. We expect this to reduce its adoption on the Internet in the future. We also show that because QUIC is implemented in the userspace, it offers a low barrier to the deployment of new and modified congestion control algorithms. In the face of this growing heterogeneity, we propose future-proof techniques for detecting known and unknown congestion control algorithms in the wild. This fresh measurement methodology was not only able to confirm BBR's reduced adoption on the Internet since 2019 but also identify undocumented congestion control algorithms being deployed by Google and Akamai on the Internet.
Overall, our 5-year study reveals that today's Internet congestion landscape is more heterogeneous than it has ever been in the past and is likely to remain this way for the foreseeable future. We expect the novel measurement methodologies presented in this seminar will help us monitor the ongoing evolution of Internet congestion control and inform the future design of modern congestion control algorithms for a heterogeneous Internet.