AI-Based Resource Management for Network Slicing in Cloud-Native Mobile Networks

The main contribution of this work is the development of an artificial intelligence (AI)-based system that dynamically manages the resources of the different slices of a 5G network core deployed in Kubernetes. The objective is to vertically scale the different resources assigned to Cloud-Native Network Functions (CNFs) without restarting the containers to maintain the session information of the 5G Core. To fulfil the requirement of hot vertical scaling, it is necessary the development of a network slicing controller which implements a vertical scaling solution using the Kubernetes API and the use of InPlaceVerticalPodScaling, a Kubernetes feature gate in beta state.Furthermore, this framework integrates two additional AIbased components to perform real-time decision-making: a Continual Learning (CL) model to predict the number of users of each slice type in the short term; and an offline Reinforcement Learning (RL) agent to prescribe optimal actions on the resources of the different slice types based on the current state of the network, in order to anticipate possible over or underutilisation situations.The work carried out in this paper provides a solution that anticipates certain undesirable situations by dynamically managing the resources of the slices, thus facilitating progress towards the implementation of Cloud-Native mobile networks.