Stepwise Optimal Inter-Slices Radio Resource Scheduling for Service-Level Agreement Assurance

Abstract

In 5G networks and beyond, radio access networks (RANs) must be able to support multiple services with different service level agreements (SLAs). Network slicing is a critical concept in this context and it depends on an efficient approach for radio resource scheduling (RRS). Inter-slices RRS is responsible for allocating resource block groups (RBGs) to the slices to ensure their SLAs. Mainly motivated by the O-RAN initiative, several works in the literature have presented proposals based on machine learning (ML) to solve this problem. However, there is still a lack of problem formalization and an optimal strategy, which are both introduced in this work. Through simulations, we compare our approach with a state-of-the-art deep reinforcement learning (DRL) agent. The results show the excess resources employed by the agent when they are plentiful, suggesting an unnecessary increase in energy consumption. Additionally, we show the relevant gap between solutions when the resources are scarce. Finally, we discuss guidelines on how to improve ML-based approaches to the inter-slices RRS problem.