Abstract

Edge computing is increasingly applied to various systems for its proximity to end-users and data sources. To facilitate the deployment of diverse edge-native applications, container technology has emerged as a favored solution due to its simplicity in development and resource management. However, deploying edge applications at scale can quickly overwhelm edge resources, potentially leading to violations of service-level objectives (SLOs). Scheduling edge containerized applications to meet SLOs while efficiently managing resources is a significant challenge. In this paper, we introduce Jingle, an autoscaler for edge clusters designed to efficiently scale edge-native applications. Jingle utilizes application performance metrics and domain-specific insights collected from IoT devices to construct a hybrid model. This hybrid model combines a predictive-reactive module with a lightweight learning model. We demonstrate Jingle’s effectiveness through a real-world deployment in a classroom setting, managing two edge-native applications across edge configurations. Our experimental results show that Jingle can fulfill SLO requirements while requiring up to 50% fewer containers than a state-of-the-art cloud scheduler, which highlights its resource management efficiency and SLO compliance.