Two-Tier Resource Allocation in Dynamic Network Slicing Paradigm with Deep Reinforcement Learning

Network slicing is treated as a key technology of the rapidly developing 5G system. Nevertheless, the environment of the users is extremely complex, leading to a great challenge for allocating the slices in an optimal manner. In this paper, we propose a dynamic slice allocation scheme with two- tier paradigm in consideration of the quality of experience (QoE). In the first tier, called local tier, we employ linear programming aided by a penalty function to allocate the radio resources in the slices to services for user equipments aiming at the best QoE. In the second tier, called edge tier, we design a deep reinforcement learning algorithm to dynamically allocate the computing resources to the edge networks, to achieve the best QoE and highest resource utilization rate. Simulation results demonstrate that the proposed paradigm can achieve better throughput and QoE in comparison with the traditional network slicing paradigms.