Dynamic Queue Management and Packet Loss Mitigation in P4-Enabled Data Planes

Modern network infrastructures, including data centers, 5 G edge nodes, and large-scale IoT networks, frequently handle dynamic and sporadic data traffic. Meanwhile, the traditional static-threshold-based congestion control often leads to excessive packet loss and inefficient resource allocation. This paper introduces a $\mathbf{P4}$-based adaptive queue management approach that dynamically adjusts congestion thresholds, leverages QoS-driven recirculation, and enables rapid fault recovery. The proposed method adapts congestion thresholds based on queue depth, selectively reprocesses high-priority packets, and generates real-time summaries when packet losses exceed critical levels. Experimental results on the BMv2 software switch, across traffic loads from 50 Mbps to 300 Mbps, demonstrate that the proposed approach reduces packet loss by $59\%$ compared to traditional static-threshold congestion control and $44.6\%$ relative to RED, while achieving a recovery rate of $67\%$ faster and a throughput $27\%$ higher than the traditional method. Although the approach incurs an additional latency of $\mathbf{1.8 ~ ms}$ over the traditional method and 1.9 ms over RED due to recirculation overhead, it significantly improves packet retention and congestion management, effectively mitigating packet loss.