Kubernetes-Based Edge System for Managing Intermittently Connected IoT Gateway on Rail Vehicle

The rail industry is shifting from time-based maintenance to condition-based maintenance (CBM) to improve maintenance efficiency and reduce maintenance costs. For CBM, IoT gateways are installed in rail vehicles to collect and pre-process sensor data before sending them to an analysis system in the cloud. Containers make it easy to flexibly update CBM’s application components in the gateway; Kubernetes is the de-facto standard for container management. However, Kubernetes is not designed for intermittently connected gateways; it causes node health issues and containers down when the gateway is disconnected. We propose an edge management system that enables the management of containers in intermittently connected IoT gateways via the Kubernetes API. The novelty of the proposed system is that it enables intermittently connected gateways managed by Kubernetes without modifying the Kubernetes control plane logic. We evaluated the system with both EKS and on-premises Kubernetes and showed that containers in a gateway managed by the proposed system successfully worked without any container downtime due to node health issues. The results showed that the edge node stayed in the Ready state and all the containers on the edge node kept running without down when the node was disconnected. Also, they showed that a container was restarted in 840 milliseconds on average if the container process was down while the edge node was disconnected. Based on these results, we concluded the self-healing mechanism for pod failure worked on the Kubernetes cluster with intermittently connected edge nodes by applying the proposed edge management system. The contribution of this paper is the architecture of the Kubernetes-based edge system for intermittently connected gateways and a feasibility test with both AWS EKS, an AWS-managed Kubernetes service, and on-premises Kubernetes.