GNN Embedded SFC Provisioning Scheme for Efficient Resources in SDN/NFV

Network function virtualization (NFV) and software-defined networking (SDN) have gained widely utilization in the ubiquitous 5G network. NFV allows virtualization network function (VNF) deployment with lower resource consumption. VNF is dedicated to sequence resources of service function chaining (SFC), which manages VNF deployments to adopt its high utilization. There are challenges with resource constraints, real-time deployments, and flexible adjustment resource pools in different situations. To tackle this problem, the graph neural network (GNN), represents all the VNF deployments as nodes, and using the message-passing graph neural network (MPGNN) method consists of feature nodes and links. MPNNs propagate the information of features in VNF to neighbor nodes, allowing each node to update its representation based on the information received from its neighbor’s nodes. After getting the graph structure of SFC. Meanwhile, it has come out with predicting the potentials of the virtual node for future in long-term feasibly changing path. We proposed Markov decision process (MDP) to solve the SFC deployment problem of facilitating the time complexity and deep reinforcement learning process to adaptively optimize network adjustments policy.