Resource Allocation for Multi-User MEC System: Machine Learning Approaches

Yong Zhang,Xiaomin Zhou,Yinglei Teng,Jinghui Fang,Weijun Zheng

DOI: https://doi.org/10.1109/csci46756.2018.00159

2018-01-01

Abstract:Mobile Edge Computing (MEC) is a promising computing paradigm, which can provide mobile devices with powerful computing capability and low delay by offloading resource-intensive tasks to the MEC Server. However, due to the varying wireless channel conditions and different task delay requirements, the resource allocation decision made by MEC Server cannot always achieve the lowest system energy consumption. In this paper, we consider resource allocation problem in a multi-user MEC system. The objective is to achieve minimum system energy consumption under delay constraint by suitably allocating communication and computing resources. We solve the communication resources (subcarrier) allocation problem through regression algorithm, make use of three regression algorithms to predict subcarrier allocation. And then calculate the computing resources (CPU frequency) allocation and system energy consumption. Simulation results show that machine learning can achieve near-optimal system energy consumption, where the system energy consumption of the ANN(Artificial Neural Network), Xgboost (Extreme Gradient Boosting), and RF(Random Forest) algorithms is 1.30%, 7.49%, and 23.29% higher than the optimal system energy consumption respectively. In the view of prediction loss, ANN and Xgboost perform better than RF.