Dynamic Computation Offloading and Resource Allocation for Multi-user Mobile Edge Computing

We study the problem of dynamic computation offloading and resource allocation in mobile edge computing (MEC) systems consisting of multiple mobile users (MUs) with stochastic task arrivals and wireless channels. Each MU can execute its task either locally or remotely in an MEC server. The objective is to identify the optimum scheduling scheme that can minimize the long-term average weighted sum of energy consumption and delay of all MUs, under the constraints of limited transmission power per MU and limited computation resources at the MEC server. The optimum design is performed with respect to three decision parameters: whether to offload a given task, how much transmission power to be allocated for offloading, and how much MEC resources to be allocated for an offloaded task. We propose to solve the problem by developing a dynamic scheduling strategy based on deep reinforcement learning (DRL) with deep deterministic policy gradient (DDPG). Simulation results show that the proposed algorithm outperforms other existing strategies such as deep Q-network (DQN).