Optimization on Service Function Chain Deployment for Cloud Network Collaborative Operation Platform

Abstract

With the involvement of Network Function Visualization (NFV), the operation cost of the cloud network collaborative operation platform can be largely abated. Whereas most service function chain (SFC) orchestration methods cannot simultaneously optimize the resource utilization while minimizing the performance of service delay. In this article, a deep reinforcement learning (DRL) based method of SFC deployment based on a cloud network collaborative operation platform is proposed. By optimizing the SFC sequence and the actual amount of resources being allocated, the proposed framework aims to minimize the resource cost while simultaneously to minimize the end to end delay of the SFC deployment. To solve the multi-objective optimization problem (MOP), a deep reinforcement learning (DRL) based framework is further explored. The MOP SFC orchestration problem is first decomposed into a group of subproblems, and each subproblem is modelled as a neural network, wherein an actor-critic algorithm and a modified pointer network are adopted to solve each subproblem. Pareto front optimal solutions can be acquired directly via the trained models. The experimental results show that the proposed method can efficiently and effectively solve the SFC deployment problem and outperform NSGA-II and MOEA/D in the aspect of solution convergence, solution diversity, and computing time. In addition, the trained model can be applied to newly encountered problems without retraining.