A MAPE-K and Queueing Theory Approach for VNF Auto-scaling in Edge Computing

Network Function Virtualization (NFV) and Edge Computing (EC) can accommodate various services on a shared virtualized infrastructure. By using distributed resources available at the network edge, the EC paradigm contributes to share service provisioning. To improve agility and flexibility for service provisioning while reducing deployment costs for infrastructure providers, NFV virtualizes computing, network, and storage resources to decouple network functions from the underlying hardware. Therefore, typical network functions in a virtualized network environment are implemented as software entities called Virtual Network Functions (VNFs), which can run on Virtual Machines (VMs) or containers within off-the-shelf servers. The integration of EC and NFV allows the creation of VNF chains, known as Service Function Chains (SFC), representing end-to-end services and their deployment on edge servers. Edge nodes tend to provide fewer stable services once the environment where they are located is unpredictable. Thus, running SFCs with an unpredictable workload is challenging, and many components may cooperate to meet the required Service Level Agreement (SLA). Therefore, such environments require strategies for automatically scaling VNFs as a function of workload fluctuation. This work addressed the VNF scaling problem by providing a novel MAPE-K-based architecture and a queue-based algorithm to dynamically scale VNF in the edge. We demonstrate that the proposed approach outperforms purely reactive auto-scaling.