Optimal ACL Policy Placement in Hybrid SDN Networks: A Reinforcement Learning Approach

The immediate transition to a fully Software-Defined Networking (SDN)-based architecture is both costly and operationally complex. As an alternative, a hybrid SDN architecture is introduced, where SDN-enabled devices are incrementally deployed alongside legacy network devices. This coexistence, however, presents significant challenges in network management and control, particularly in the efficient implementation of Access Control List (ACL) policies. ACL policies play a crucial role in defining network security and control mechanisms, yet existing approaches assume that all subnets affected by these policies are continuously transmitting data. Given the dynamic nature of subnet behavior, such static assumptions can lead to suboptimal network performance. To address this limitation, this paper proposes a novel Q-learning-based approach for dynamically deploying ACL policies based on real-time data transmission patterns of subnets. The proposed algorithm optimally determines the placement of ACL policies, minimizing both the total number of policies and redundant transmissions in the network. Extensive evaluations using real-world network traces and topologies demonstrate that our approach outperforms existing state-of-the-art methods in terms of efficiency and network performance.