Building the Digital Twin of a MEC node: a Data Driven Approach

Abstract

Multi-access edge computing (MEC) represents an emerging solution to improve the performance of mobile networks by bringing computing resources closer to the edge of the network. However, MEC requires the implementation of virtualization and can be deployed using different hardware platforms, including COTS devices. In this highly heterogeneous scenario, the digital twin (DT), assisted by proper AI/ML solutions, is envisioned to play a crucial role in automated network management, operating as an intermediate and collaborative layer enabling the orchestration layer to better understand network behavior before making changes to the physical network. In this paper, we aim to develop a DT model that captures the behavior of a MEC node supporting services with varying workloads. In pursuit of this objective, we adopt a data-driven methodology that effectively learn a model predicting three critical key performance indicators (KPIs): throughput, computational load, and power consumption. To demonstrate the viability and potential of such approach, a measurement campaign is conducted on MEC nodes deployed with different virtualization environments (bare metal, virtual machine, and containerized), and the results are used to build the DT of each node. Furthermore, machine learning models, including k-nearest neighbors (KNN), support vector regression (SVR), and polynomial fitting (PF), are used to understand the amount of actual measurements required to achieve a suitably low KPI prediction error. The results of this study provide a basis for further research in the field of MEC DT models and carbon footprint-aware orchestration.