Beyond bufferbloat: End-to-end congestion control cannot avoid latency spikes

End-to-end congestion control methods like TCP and QUIC are crucial for managing internet congestion, but a recent paper by Domos highlights a key limitation—they cannot prevent latency spikes. Despite efforts to improve TCP's latency performance, issues persist, especially in networks with rapidly changing link capacities like Wi-Fi and 5G. Bufferbloat, identified as a major source of latency, results from poor congestion signaling causing unnecessary queuing. While progress has been made in reducing bufferbloat, challenges remain in achieving low latency. The paper emphasizes that existing congestion control methods cannot fully address latency spikes, necessitating alternative solutions. Suggestions include anticipating capacity changes, underutilizing links, prioritizing latency-sensitive traffic, and diversifying links to enhance reliability. The study underscores the need to acknowledge and address these limitations to realize a dependable low-latency internet, particularly in the context of 5G networks. Researchers and engineers are encouraged to explore these alternative strategies to mitigate latency issues effectively.