AI-Based Performance Enhancement for Multi-Tenant Slicing

With the maturity of 5G standards and the commercialization of 5G, more and more vertical industries are brought in the business model of mobile network, opening new market opportunities to Mobile Network Operators (MNOs). Vertical industries as tenants provide customized services by renting resources from MNOs to deploy slices. Network slicing enables these tenants to efficiently share resource of the same federated infrastructure network. However, as the number of tenants with differentiated requirements increases, resource competition among the slices owned by different tenants will bring the degradation of slice traffic performance. This chapter first introduces the collaborative business model of multi-tenant slicing and illustrates the resource competition using a multilayer network model. After obtaining a global perspective of resource utilization of multi-tenant slicing, traffic performance analysis of multiple isolated slices are conducted with the slice traffic model. The influence of different factors, such as the number and scale of slices, nodal coverage and resource allocation parameters, is analyzed comprehensively. Based on the analysis results, this chapter proposes control strategies for avoiding the damage of resource competition on slice traffic performance. To deal with the dynamics and complexity in resource management of multi-tenant slicing, the architecture and advantages of three newest deep reinforcement learning (DRL) algorithms are presented. Applying these algorithms in multi-tenant slicing can achieve faster convergence speed and better performance.