Horizontal Auto-Scaling in Edge Computing Environment using Online Machine Learning

A major challenge in edge computing platforms for container-based applications is dealing with the dynamic workload. At certain times, the resources previously allocated to a given container may not be adequate for a substantial increase in processing requests. This paper proposes an online machine learning auto-scaling approach for applications running at the network edge. The auto-scaling follows the MAPE-K control loop architectural style to dynamically adjust the number of containers in response to workload changes. The approach was designed for scenarios at the edge of the network where the behavior of the data is unknown beforehand. We designed a hybrid auto-scaling mechanism that behaves reactively while a prediction online machine learning model is continuously trained. When the prediction model reaches a desirable performance, the auto-scaling behaves proactively using predictions to anticipate scaling actions. The evaluations have demonstrated the feasibility of the architecture. Compared with proactive and reactive autoscaling, it achieved the best results, decreasing the scaling actions, preventing insufficient provision situations, and assuring the meeting of QoS requirements.