Service Deployment and Migration Methods Based on RNN in MEC-Enabled HetNet

With the increasing number of 5G user equipment (UEs), the need for real-time data processing is growing. However, the constrained computing power and battery capacity of UEs cannot meet all computing requirements. Traditional cloud computing has been able to solve this problem in the past, but the transmission latency is too high to meet the 5 G UEs’ requirements for lower delay. Recently, Multi-Access Edge Computing (MEC) has become a promising research topic, and a MEC Heterogeneous Network (HetNet) can contain cells of different sizes, with each cell equipped with a MEC server. However, when the UEs leave the original MEC server, service interruptions occur, which leads us to this proposal. We first design a Recurrent Neural Network (RNN)-based service deployment and migration method to tackle this problem, then further design a Long Short-Term Memory (LSTM)-based method and a Gated Recurrent Unit (GRU)-based method for improving performance. Experimental results verify that our methods outperform other existing methods on energy consumption, service completion time, fairness and prediction error rate of tasks and resource utilization of MEC servers in different scenarios.