Reinforcement Learning-Based Resource Allocation in Edge Computing

摘要

The problem of online resource allocation in edge computing has become a research hotspot. Meanwhile, reinforcement learning (RL) is suitable for solving online problems. In this paper, we combine edge computing online resource allocation with RL. This combination enables edge computing resource providers to obtain more social welfare and improve resource utilization. Specifically, we define a dynamic resource allocation problem for edge computing: edge equipment users request resources from a nearby edge computing server (ECS), the amount of resources required varies among the users, and there are time limits for the completion of the requested tasks. Since this resource allocation problem is NP-hard, it cannot be solved in polynomial time. Therefore, we propose an algorithm based on the policy-gradient algorithm in RL to solve the problem. Our approach is experimentally compared with existing research in terms of social welfare and resource utilization, for which it achieves good results.