Service Migration with High-Order MDP in Mobile Edge Computing

Abstract

Mobile Edge Computing (MEC) has gained a lot of popularity for supporting newly emerging services with low latency by deploying servers in base stations (BSs). For a moving user, the Quality of Service (QoS) needs to be guaranteed through service migration among edge servers. To avoid service interruption, determining when and where to migrate services is critical and challenging for users with unknown future trajectories. Besides, frequent service migration incurs unexpected high network resource consumption, resulting in a trade-off between QoS and resource cost. In this paper, we focus on the problem of service migration in MEC networks aiming at minimizing the total resource cost while guaranteeing the QoS of moving users. We innovatively model the service migration of a user using k-order Markov Decision Process (MDP), where the correlation of user’s historical locations is taken in account to help better decision. The optimal correlation coefficient k is obtained through analyzing the real-world dataset of taxi trajectories. An online algorithm based on Deep Q-Network (DQN) is proposed to solve the service migration problem to minimize the long-term communication and migration costs. Compared with the widely-used benchmarks, our algorithm shows a better performance in reducing communication and migration costs under different parameter settings.