Performance Modeling of Heterogeneous Edge-Cloud Systems with Machine Learning

Edge-cloud systems are heterogeneous computational infrastructures designed to manage distributed workloads efficiently. Accurate prediction of system performance is essential for minimizing execution time, energy, and enhancing throughput. Traditional performance evaluation methods, such as simulation-based tools, are often time-consuming, require manual configuration, and rely heavily on assumptions that limit scalability and adaptability. To address these challenges, this work investigates several machine learning (ML) models to predict performance in heterogeneous edge-cloud environments, focusing on key metrics such as execution time, energy consumption, and throughput. Five different ML models, namely, Random Forest (RF), Long Short-Term Memory (LSTM), Deep Neural Network (DNN), Recurrent Neural Network (RNN), and a hybrid RNN-DNN, are developed and evaluated. The training dataset is generated using the VisualSim system-level modeling tool by simulating diverse edge-cloud configurations. The performance of the models is evaluated using the mean absolute error (MAE) and the root mean square error (RMSE), and the predicted results are validated against the VisualSim outputs. Experimental results show that the RF model achieves the lowest MAE and RMSE on the test datasets. The deep learning models exhibit varying levels of accuracy, with the DNN model offering a strong trade-off between computational complexity and predictive performance.