Multi-armed Bandit Algorithm for Online Offloading and Scheduling in Edge Computing Environment

In this paper, we study the online scheduling offloading problem in edge computing. Previous scheduling work generally assumes that system have global information, or between the server and the user mutual state information known. Considering the privacy security and the conflicts that users may have during wireless communication, we designed a stochastic multi-played multi-armed bandit online learning offloading framework: D-UCB-G. The framework performs offloading and scheduling based on historical experience, which maximizes the service quality of user devices and also solves the problem that users cannot access the state of edge servers. In the case of ensuring that the user devices do not conflict, the user devices are scheduled to use the edge server resources. We prove the convergence of the algorithm experimentally and demonstrate the effectiveness of the algorithm by applying real-world location data to simulations.