A Multi-Armed Bandit Game for Multi-Tenant RAN Slicing

Network slicing is an auspicious technology in 5G networks that can handle different scenarios and use cases. Resource scheduling is essential for improving resource-multiplexing gain among slices while meeting specific service requirements for Radio Access Network (RAN) slicing. It needs to be reconfigured adaptively to reduce re-slicing frequency, maximize network resource utilization, and guarantee performance and isolation between multi-tenants under dynamic traffic load. However, the resource reconfiguration problem is intractable due to the high computational complexity caused by the numerous variables. Thus, this paper proposes an intelligent resource reconfiguration for multi-tenant RAN slicing. We explore a collaborative online learning framework consisting of multi-armed bandits (MAB) for tackling small-time-scale resource allocation and propose the Exp3 algorithm. Based on the historical traffic data, our approach assists the RAN in making online resource re-scheduling decisions. Extensive simulations validate our proposed Learning-Based approach’s effectiveness compared with other benchmark algorithms (e.g., Threshold-Based, Periodic slicing). In addition, we investigate the Exp3 algorithm under different parameter configurations. Numerical results show the convergence of our proposed model, improve operator satisfaction and resource utilization, and reduce the frequency of re-slicing by ensuring multi-service and multi-tenant isolation.