Optimal Resource Allocation in SDN/NFV-Enabled Networks via Deep Reinforcement Learning

Software-Defined Networking (SDN) and Network Functions Virtualization (NFV) are two emerging paradigms that enable the feasible and scalable deployment of Virtual Network Functions (VNFs) in commercial-off-the-shelf (COTS) devices, which deliver a range of network services with reduced cost. The deployment of these services requires efficient resource allocation that fulfills the requirements in terms of Quality of Service (QoS) and Service-Level Agreement (SLA) while considering the constraints of the underlying infrastructure, such as maximum latency tolerance and affinity policies. To address this issue, we study the resource allocation problem in SDN/NFV-enabled networks, which involves numerous optimization variables resulting from the multidimensional space of system component parameters and states. Using deep reinforcement learning, we propose a policy gradient-based algorithm with an invalid action masking approach to efficiently tackle the resources allocation problem while handling system constraints in industrial settings. The simulation results unequivocally show the effectiveness and performance of the proposed learning approach for this category of problems.