Deep Reinforcement Learning with Graph Neural Networks for Capacitated Shortest Path Tour based Service Chaining

Network functions virtualization (NFV) realizes diverse and flexible network services by executing network functions on generic hardware as virtual network functions (VNFs). A certain network service is regarded as a sequence of VNFs, called service chain. The service chaining (SC) problem aims at finding an appropriate service path from an origin node to a destination node while executing the VNFs at the intermediate nodes in the required order under resource constraints on nodes and links. The SC problem belongs to the complexity class NP-hard. In our previous work, we modeled the SC problem as an integer linear program (ILP) based on the capacitated shortest path tour problem (CSPTP) where the CSPTP is an extended version of the SPTP with the node and link capacity constraints. We also developed the Lagrangian heuristics to achieve the balance between optimality and computational complexity. In this paper, we further propose a deep reinforcement learning (DRL) framework with the graph neural network (GNN) to realize the CSPTP-based SC adaptive to changes in service demand and/or network topology. Numerical results show that (1) the proposed framework achieves almost the same optimality as the ILP for the CSPTP-based SC and (2) it also works well without retraining even when the service demand changes or the network is partly damaged.