Deep Reinforcement Learning-based Task Offloading and Resource Allocation in MEC-enabled Wireless Networks

Mobile edge computing (MEC) has recently become an enabling technology for mobile operators that are offering a diverse set of services. These services require extensive storage, energy, and computation resources. However, user devices (UDs) have resource constraint to meet the requirements of such services. To tackle the contradiction between resource-constrained UDs and computationally intensive services, MEC have been proposed. MEC servers provide task execution services for UDs. On receiving a service request from the UDs, a MEC server within the network may dynamically allocate computation and memory resources for the task execution. As the MEC servers have limited capacity, efficient utilization of MEC resources is necessitated. Also, it is challenging to find an optimal solution for efficient resource allocations due to different task requirements for a diverse set of services offered to users and dynamicity in wireless networks. To address these problems, we propose a partial task offloading and resource allocation scheme to maximize user task completion within a tolerable time period while minimizing energy consumption. In this paper, we convert the formulated optimization problem to a markov decision process (MDP) and then propose a solution based on the deep deterministic policy gradient (DDPG) algorithm. The performance results show that the proposed method completes a greater number of tasks within a tolerable delay and reduces the energy consumption in the network, compared to those of other conventional schemes.