Published
Edge Kubernetes: CPU & Energy
Energy comparison of five lightweight Kubernetes distributions (k8s, k3s, k0s, KubeEdge, OpenYurt) on ARM edge hardware via a DVFS power model, with interrupt and thermal analysis — finding CPU frequency does not predict orchestration performance and k8s is most efficient per pod. Third in the edge series; ESOCC 2026, Springer LNCS
About
The third study in the lightweight-Kubernetes edge series (after the two ESOCC 2025 chapters), and the first to model energy rather than raw CPU. A DVFS-based power model calibrated to the ARM Cortex-A72 (Raspberry Pi 4) compares five distributions — k8s, k3s, k0s, KubeEdge, OpenYurt — across idle and pod-churn workloads, with interrupt-storm profiling and thermal analysis. At idle all five draw within ~1.5%; under load k8s is the most energy-efficient per pod lifecycle despite being full-featured, and CPU frequency does not predict orchestration performance — the differences are software-architectural, not hardware. Full title "Comparative Analysis of Lightweight Kubernetes Distributions for Edge Computing: CPU Microarchitecture and Energy Efficiency", published in ESOCC 2026 (Springer LNCS) with Sergio Moreschini and David Hästbacka.