A Multi-Agent Based Dynamic Network Slice Tarification Framework
Résumé
5G networks promise to satisfy diverse kinds of advanced use cases, by providing tailored services to different kinds of customers, through the concept of Network Slicing (NS). Although NS, i.e. multiple virtual networks running on a shared infrastructure, is expected to offer substantial advantages in terms of flexibility and cost-efficiency, there are several key issues that have yet to be addressed. One of the open challenges that have emerged is regarding the economic interaction between Mobile Network Operators (MNOs) and the multitude of tenants (e.g. vertical industries) that MNOs intend to serve. With the paradigm shift from a product-based to a service-based model, MNOs need to come up with sophisticated pricing strategies for their heterogeneous set of customers. To address this issue, we propose a dynamic slice tarification algorithm based on Reinforcement Learning (RL) that adapts slice prices in accordance with the microeconomic laws of supply and demand. To model the system and prove the performance of our approach, we use a Multi-Agent Simulation (MAS) of our NS scenario. We train and evaluate our approach on a monopolistic scenario with one MNO and a number of tenants.