Performance Analysis of Access and Mobility Management Function On a 5G Core Based On CPU Usage Predictions

Abstract

The increasing number of mobile devices and the growing demand for services lead to an increase in the access requests per second to the Access and Mobility Management Function (AMF) of the control plane in a Fifth Generation (5G) mobile network. It causes congestion of the function and affects the overall network performance. Therefore, this paper proposes a self-scaling mechanism for the AMF in a 5G core by CPU usage predictions using the Long Short-Term Memory (LSTM) machine learning (ML) technique. The mechanism predicts the percentage of CPU usage in the pod containing the AMF and establishes scaling policies that determine the necessary number of AMF pods. The performance of the AMF is evaluated through success rate, loss rate, and latency of access requests per second in three scenarios: a reactive one with scaling based on current CPU thresholds, a predictive one using CPU predictions, and another using both the scaling policies and the LSTM technique. With the previous scenarios, the AMF is scaled reactively and predictively. Results show that the scaling policies and the ML algorithm significantly improve the performance of the function in terms of success rate and loss rate of access requests per second. An efficient self-scaling of the AMF is achieved, which contributes both to the optimization of computational resources and to improving the availability of the 5G mobile network.