Abstract: Artificial intelligence (AI)-native communication is considered one of the key technologies for the development of 6G mobile communication networks. This paper investigates the architecture for developing the Network Data Analytics Function (NWDAF) in 6G AI-native networks. The architecture integrates two key components: data collection and management, and model training and management. It achieves real-time data collection and management, establishing a complete workflow encompassing AI model training, deployment, and intelligent decision-making. The architecture workflow is evaluated through a vertical scaling use case by constructing an AI-native network testbed on Kubernetes. Within this proposed NWDAF, several machine learning (ML) models are trained to make vertical scaling decisions for User Plane Function (UPF) instances based on data collected from various network functions (NFs). These decisions are executed through the Kubernetes API, which dynamically allocates appropriate resources to UPF instances. The experimental results show that all implemented models demonstrate satisfactory predictive capabilities. Moreover, compared with the threshold-based method in Kubernetes, all models show a significant advantage in response time. This study not only introduces a novel AI-native NWDAF architecture but also demonstrates the potential of AI models to significantly improve network management and resource scaling in 6G networks.

Keywords: 6G; AI-native; NWDAF; UPF Scaling