Stateful Virtual Network Function Decomposition and Deployment With Reliability Guarantee in Edge Networks

Edge Networks (ENs) are emerging networks that enable deploying multiple virtual network functions (VNFs) on resource-limited edge servers to provide users with tailored virtual network services. Decomposing a single VNF into multiple thinner replicas can enhance service reliability while inevitably incurring additional computing capacity consumption (e.g., operating system overhead caused by instantiating more replicas), which increases with the number of decomposed replicas. Moreover, redundant backup replicas can be deployed near the replicas to enhance the reliability further. However, the stateful nature of VNFs requires state synchronization among replicas and between replicas and backup replicas, resulting in additional communication traffic. In this paper, we consider a joint strategy for the decomposition and deployment of stateful VNFs with the goal of minimizing total cost while meeting users’ reliability requirements. The total cost includes the computing cost for instantiating replicas and backup replicas, the additional consumption of computing capacity due to VNF decomposition, and the communication cost for routing traffic among users, replicas, and backup replicas. We first formulate the cost minimization problem as an integer nonlinear program and prove that it is NP-hard. Then, we propose an online two-stage scheme to solve this problem, where the first stage is a VNF decomposition algorithm, and the second stage is a deployment algorithm based on deep reinforcement learning (DRL). The former effectively reduces computing cost by iteratively adjusting the number of replicas and backup replicas, while aiding the latter to adaptively minimize communication cost. Extensive experiments demonstrate that our scheme is promising compared to existing state-of-the-art methods.