Joint Optimization of Resource Allocation and Service Performance in vEPC Using Reinforcement Learning

Network Function Virtualization (NFV) is the transition from proprietary hardware functions to virtualized counterparts of them within the telecommunication industry. With the aim of quality of service guarantee and energy saving, telco operates need to decided when and how to scale the virtual resource with the traffic processing demand. In this paper, we proposed an auto-scaling mechanism based on reinforcement learning. First, we establish a system model for vEPC (virtualized Evolved Packed Core) to gather the state information. Second, auto-scaling mechanism based on reinforcement learning can treat procedure of scaling decision as Markov Decision Process. By simulation, our mechanism outperforms threshold based policy, and realizes the joint optimization of resource allocation and service performance in vEPC.