Workload Prediction of Virtualized RAN in the Edge Micro Data Center: An Experimental Progress

This paper introduces the importance of workload prediction for virtualized, disaggregated radio access network (RAN) deployment in the edge micro data center (EMDC). To predict the workload, several machine learning algorithms, e.g., ARIMA and LSTM, are evaluated in terms of central processing unit (CPU) usage from the Kubernetes cluster while deploying the 5G vRAN components (i.e., radio unit, distributed unit, and centralized unit) in the EMDC. In addition, we have validated the prediction results by using data collected from an experimental testbed. Our investigation demonstrates that the LSTM model offers a practical advantage in implementing and utilizing it within the EMDC context, without the need for extra computational resources when compared to utilizing transfer learning with ARIMA.