Proactive Deployment of Chain-based VNF Backup at the Edge using Online Bandit Learning

The emergence of mobile edge computing (MEC) empowers the popularity of resource-consuming applications on edge devices, where virtual network functions (VNFs) are widely employed to provide flexible and scalable network service to users in an individual or chained manner. To maintain high availability of VNFs, deploying redundant backups of VNFs on edge servers has become an effective method for swift recovery. However, the existing studies largely ignore the impact of heterogeneous VNF importance on appropriately placing VNF backups at the edge. In this work, we specify the concept of VNF demand levels considering both the user service needs and service function chain (SFC) composition requirements to fill the above gap, which is employed as a significant factor to formulate the edge-assisted VNF backup issue as an optimization problem with the resource constraint of edge servers. To solve the challenge brought by the uncertain demand levels of VNFs, we resort to the combinatorial multi-armed bandit (CMAB) problem to propose an online learning based approximate VNF backup selection and deployment algorithm, named BSPS, with acceptable computation complexity and regret bound. Furthermore, regarding the balance of workload among edge servers, the basic problem is extended to accommodate the extra constraint and an additional VNF backup scheme is devised accordingly. Extensive simulation results with numerical analysis demonstrate the attractive performances of our proposed schemes.