Key points are not available for this paper at this time.
Mobile Edge Computing (MEC) is a technology that provides computing services to mobile users by deploying edge servers at base stations, and it has become an important tool for enhancing user service experience. Due to limited resources and user mobility, service migration among edge servers becomes an effective way to maintain user service latency. However, frequent service migrations and resource competition among users result in additional energy consumption, significantly increasing the operational costs of service providers. In complex dynamic and resource-competitive scenarios, considering migration energy consumption as an independent core optimization objective comparable to service latency and making adaptive online migration decisions pose significant challenges and remain underexplored. In this paper, we propose a systematic new approach to address the service migration challenges in MEC from the perspectives of theoretical modeling, algorithm design, and mechanism innovation, namely a delay- and energy-aware adaptive service migration method. Firstly, we model service migration as a constrained (resource competition) bi-objective optimization problem (minimizing service latency and migration energy consumption), which can be formulated as a mixed-integer nonlinear programming problem. Then, given the NP-hard nature of this problem, network dynamics, differences in user mobility patterns, and the real-time requirements of migration decisions, we design a migration model based on Deep Q-Network (DQN), referred to as NPER-D3QN. This model integrates elements of D3QN, Noisy Net, and prioritized experience replay, enabling the generation of adaptive migration strategies with low latency and low energy consumption according to real-time network states, user mobility patterns, and server loads. Finally, we conduct simulation experiments comparing our proposed service migration method with existing typical and advanced methods. The results demonstrate that our method exhibits superior performance in terms of latency and energy efficiency, showing certain advantages in dynamic adaptation and multi-objective collaborative optimization.
Li et al. (Sat,) studied this question.