Anomaly Detection-Based Resource Autoscaling Mechanism for Fog Computing

Fog computing leverages the cloud computing services at the network edge. It aims to address the issues of latency, bandwidth constraints, and network reliability in the context of IoT. However, guaranteeing Service Level Agreements (SLAs) while adapting to the dynamicity of the requests raised by the IoT devices and the edge applications and ensuring efficient performance is challenging in fog computing environments. Resource autoscaling in response to dynamic workloads is one of the approaches to overcome the aforementioned challenges. The proposed work introduces an anomaly detection-based resource autoscaling mechanism based on the deep autoencoders. Deep autoencoders use neural networks to detect the anomalies in the data, based on which Virtual Machine (VM) autoscaling within the fog nodes can be facilitated. Anomaly scores and response times for the three variants of deep autoencoders are compared. The outcome indicates that the proposed approach improves scalability, responsiveness, and resource utilization.