Performance Evaluation of Kubernetes Networking Approaches across Constraint Edge Environments
Georgios Koukis, Sotiris Skaperas, Ioanna Angeliki Kapetanidou, Lefteris Mamatas, Vassilis Tsaoussidis
TL;DR
This work evaluates how Kubernetes networking performance at the edge is affected by the interplay between CNI plugins and lightweight Kubernetes distributions. It combines qualitative plugin/distribution profiling with quantitative measurements of CPU, RAM, and throughput across intra- and inter-host scenarios using two edge testbeds and knb benchmarking. Key findings show that while lightweight distributions can reduce RAM for some plugins, they do not universally improve resource usage or throughput, and certain plugins (e.g., Calico, Kube-router) exhibit distribution-dependent performance trade-offs. The results inform edge practitioners about selecting combinations of distributions and CNIs to balance resource constraints with networking performance, and point to further optimization opportunities and broader metric studies for edge scenarios.
Abstract
Kubernetes (K8s) serves as a mature orchestration system for the seamless deployment and management of containerized applications spanning across cloud and edge environments. Since high-performance connectivity and minimal resource utilization become critical factors as we approach the edge, evaluating the performance of K8s networking in this context is essential. This paper contributes to this effort, by conducting a qualitative and quantitative performance evaluation of diverse Container Network Interface (CNI) plugins within different K8s environments, incorporating lightweight implementations designed for the Edge. Our experimental assessment was conducted in two distinct (intra- and inter-host) scenarios, revealing interesting insights for both researchers and practitioners. For example, the deployment of plugins across lightweight distributions does not necessarily lead to resource utilization improvements, e.g., in terms of CPU/memory or throughput.