A Mobility-Aware Service Function Chain Migration Strategy Based on Deep Reinforcement Learning

With the development of network function virtualization (NFV) and mobile edge computing (MEC), service function chaining (SFC) can be deployed more flexibly on the network edge to provide users with higher quality services. However, the mobility of users might affect the quality of the perceived service and even cause service unavailability. It is thus a challenge to better migrate the service function chains to reduce prevent such degradation and improve the migration success rate. This paper investigates the SFC migration timing decision problem in user predicted movement scenarios with a known mobile path and predicted arrival time. First, we establish an prediction model of user arrival time and formulate the SFC migration process as a mathematical model. Then, we model the SFC migration process as a Markov decision process, and propose a deep Q-network based SFC migration timing decision (DQN-MTD) algorithm. DQN-MTD can be useful to perceive and predict the state changes of network resources, and select appropriate migration timing for virtual network functions (VNFs) based on SFC migration information. The experimental results show that compared with existing algorithms, DQN-MTD algorithm can reduce the average service downtime by about 14%, improve the migration success rate of SFC by about 20%, and reduce the average VNF migration time and memory when network load is low.