Programmable Data Plane Mediation for Heterogeneous Congestion Control Ecosystems

Kotoku et al. (2025) address fairness issues when BBR and CUBIC flows coexist by classifying and relaying flows differently at intermediate devices. However, current approaches are rudimentary—typically applying static policies or simple discrimination. This research proposes an extensible, AI-powered programmable data plane (e.g., via P4) that identifies not only the congestion control algorithm in use but also the real-time context (e.g., flow age, path congestion, predicted QoS impact). The switch can then dynamically apply fairness policies, drop policies, or even “nudge” endpoints via explicit feedback. This creates a new layer of mediation that is both application- and flow-aware, potentially smoothing out incompatibilities as new congestion control algorithms proliferate. It’s particularly timely given the increasing diversity in transport protocols, and leverages recent advances in programmable networks and in-network ML.

References:

1. Programmable Data Plane Approach for Fairness Improvement Using AI-Based Classification of Congestion Control Algorithms. Yuya Kotoku, Daiki Nobayashi, Takeshi Ikenaga (2025). International Conference on Ubiquitous and Future Networks.