A PSO-SVR Based Resource Allocation Approach for Network Function Virtualization

Network Function Virtualization (NFV) provides data centers with the required programmability, flexibility, and manageability. Resource allocation of Virtual Network Functions (VNFs) is a significant challenge to deploying NFV data centers effectively. This paper proposes a novel NFV resource allocation scheme that relies on a prediction strategy of CPU utilization due to VNF traffic. The proposed scheme uses Support Vector Machine Regression (SVR) to predict CPU utilization and Particle Swarm Optimization (PSO) to optimize the critical SVR’s parameters. Experimental results on a real scenario dataset of VNFs applied in data centers show that the proposed PSO-SVR model yields better CPU prediction performance based on dynamic resource consumption by VNFs than other previously proposed models in the state-of-the-art.