DDPG-pFlow: DNS Query Prioritization and Optimization with Deep Reinforcement Learning

This paper presents a novel framework for optimizing DNS query handling through dynamic prioritization and resource management using reinforcement learning (RL). The proposed DDPG-pFlow approach leverages both content-level and context-level features to intelligently prioritize queries, ensuring efficient processing under varying network conditions, such as high load or mixed traffic. The framework incorporates the Deep Deterministic Policy Gradient (DDPG) algorithm to adaptively adjust priorities, minimizing latency and maximizing throughput. Extensive performance evaluations demonstrate that DDPG-pFlow outperforms traditional DNS query handling methods, achieving up to 70% faster query response times, 50% higher prioritization accuracy, and 45% better cache hit rates compared to Fixed Priority (FP) and other state-of-the-art methods. Additionally, it achieves 30%–40% higher throughput, up to 90% lower query drop rates, and 20%–25% better malicious traffic mitigation. The results highlight the proposed framework's effectiveness in both normal and stressful conditions, offering a scalable, robust solution for modern DNS infrastructures.