Comparative Study of Latency-Aware Serverless Function Orchestration using XGBoost and Random Forest Regression Models in Edge– Cloud Environments

Abstract: The number of Internet of Things (IoT) applications and edge–cloud computing environments has expanded, and efficient and low-latency execution of tasks is desired. Cloud native orchestration solutions are often not well suited to deal with real-time needs due to the frequent movement of cloud workloads, changes in network and resource variations. In this study, we are interested in using two ensemble ML algorithms (Random Forest and XGBoost) to predict the time required for the execution of the tasks in the vmCloud workload dataset in a serverless edge–cloud environment. Seven workload attributes—CPU utilisation, memory usage, network latency, task size, cold-start delay, number of instances, and bandwidth—were used to train the models. RMSE, MAE, MAPE, R2 score and 5-fold cross-validation were used for the performance evaluation. The results of the experiments indicate that the performance of XGBoost is better than that of the RF, with Test R2 of 0.9209, RMSE is 10.08 ms, and MAPE is 14.57%, while RF’s Test R2 is 0.8898, RMSE is 11.89 ms, and MAPE is 18.38%. The feature importance analysis shows that network latency and cold start delay are the most important features that affect the time to execute. The findings indicate that XGBoost is a proper algorithm for latency-aware serverless function orchestration and can be applied to intelligently schedule in the edge– cloud computing environment.

Keywords: Edge Computing, Cloud Computing, IoT, Serverless Computing, XGBoost, Random Forest, Function Orchestration, Latency Prediction.