EIS: Edge Information-Aware Scheduler for Containerized IoT Applications

Abstract

Edge computing has emerged as a powerful paradigm for Internet of Things (IoT) applications as it can provide computing and network services in close proximity to end devices. In an edge environment, leveraging container technology to package IoT applications offers significant benefits of flexibility and agility, while the incorporation of Kubernetes can effectively orchestrate large-scale containerized applications. However, the existing Kubernetes scheduling solutions mostly cannot satisfy IoT applications with stringent and diverse network, computing, and storage requirements, and they also lack the scalability to customize scheduling strategies. To address these, we develop an edge information-aware scheduler (EIS) based on the novel Kubernetes scheduling framework. EIS schedules containerized IoT applications by sensing the network topology and performance information of edge clusters. Moreover, EIS can make scheduling decisions according to application characteristics and resource requirements. By adopting a plug-in architecture, EIS not only provides an extensible programming interface, but is also compatible with Kubernetes’ default scheduler. We evaluate EIS in a real-world experimental environment, and the results show that EIS can reduce network latency by 18%, improve computing performance up to 140% and improve I/O performance up to 130%. These improvements are critical for IoT applications to provide high quality of service.