Unveiling Kubernetes CNI: An In-Depth Analysis of Networking Performance and Resource Efficiency

Wonmi Choi; Juyoung Ahn; Yeonho Yoo; Zhixiong Niu; Gyeongsik Yang; Chuck Yoo

doi:10.7840/kics.2025.50.7.1085

Unveiling Kubernetes CNI: An In-Depth Analysis of Networking Performance and Resource Efficiency

Wonmi Choi
, Juyoung Ahn
, Yeonho Yoo
, Zhixiong Niu
, Gyeongsik Yang
, Chuck Yoo

Department of Computer Science and Artificial Intelligence

Korea University
Microsoft USA

Research output: Contribution to journal › Article › peer-review

Abstract

Kubernetes relies heavily on its networking performance. It offers four representative networking plugins: Flannel, Calico, Cilium, and Kube-router. However, their performance differences are not well understood. This study evaluates these plugins using real-world workloads like Memcached, Nginx and Kafka, examining throughput, latency, and CPU usage in 10 GbE and 100 GbE environments. Results reveal significant performance differences due to each plugin’s architecture. Kube-router excels in CPU- and network-intensive scenarios but complicates network management. Among overlay plugins Flannel performs best in CPU-intensive tasks, while Cilium is superior for network-intensive tasks. This analysis provides insights for selecting suitable plugin based on workload characteristics.

Original language	English
Pages (from-to)	1085-1099
Number of pages	15
Journal	Journal of Korean Institute of Communications and Information Sciences
Volume	50
Issue number	7
DOIs	https://doi.org/10.7840/kics.2025.50.7.1085
State	Published - 1 Jul 2025

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

SDG 8 Decent Work and Economic Growth
SDG 12 Responsible Consumption and Production

Keywords

Cloud networking
Container network interface
CPU usage
Kubernetes
Performance profiling

Access to Document

10.7840/kics.2025.50.7.1085

Cite this

@article{10b5956b83b5470a905e6b82954c2c57,

title = "Unveiling Kubernetes CNI: An In-Depth Analysis of Networking Performance and Resource Efficiency",

abstract = "Kubernetes relies heavily on its networking performance. It offers four representative networking plugins: Flannel, Calico, Cilium, and Kube-router. However, their performance differences are not well understood. This study evaluates these plugins using real-world workloads like Memcached, Nginx and Kafka, examining throughput, latency, and CPU usage in 10 GbE and 100 GbE environments. Results reveal significant performance differences due to each plugin{\textquoteright}s architecture. Kube-router excels in CPU- and network-intensive scenarios but complicates network management. Among overlay plugins Flannel performs best in CPU-intensive tasks, while Cilium is superior for network-intensive tasks. This analysis provides insights for selecting suitable plugin based on workload characteristics.",

keywords = "Cloud networking, Container network interface, CPU usage, Kubernetes, Performance profiling",

author = "Wonmi Choi and Juyoung Ahn and Yeonho Yoo and Zhixiong Niu and Gyeongsik Yang and Chuck Yoo",

year = "2025",

month = jul,

day = "1",

doi = "10.7840/kics.2025.50.7.1085",

language = "English",

volume = "50",

pages = "1085--1099",

journal = "Journal of Korean Institute of Communications and Information Sciences",

issn = "1226-4717",

publisher = "Korean Institute of Communications and Information Sciences",

number = "7",

}

TY - JOUR

T1 - Unveiling Kubernetes CNI

T2 - An In-Depth Analysis of Networking Performance and Resource Efficiency

AU - Choi, Wonmi

AU - Ahn, Juyoung

AU - Yoo, Yeonho

AU - Niu, Zhixiong

AU - Yang, Gyeongsik

AU - Yoo, Chuck

PY - 2025/7/1

Y1 - 2025/7/1

N2 - Kubernetes relies heavily on its networking performance. It offers four representative networking plugins: Flannel, Calico, Cilium, and Kube-router. However, their performance differences are not well understood. This study evaluates these plugins using real-world workloads like Memcached, Nginx and Kafka, examining throughput, latency, and CPU usage in 10 GbE and 100 GbE environments. Results reveal significant performance differences due to each plugin’s architecture. Kube-router excels in CPU- and network-intensive scenarios but complicates network management. Among overlay plugins Flannel performs best in CPU-intensive tasks, while Cilium is superior for network-intensive tasks. This analysis provides insights for selecting suitable plugin based on workload characteristics.

AB - Kubernetes relies heavily on its networking performance. It offers four representative networking plugins: Flannel, Calico, Cilium, and Kube-router. However, their performance differences are not well understood. This study evaluates these plugins using real-world workloads like Memcached, Nginx and Kafka, examining throughput, latency, and CPU usage in 10 GbE and 100 GbE environments. Results reveal significant performance differences due to each plugin’s architecture. Kube-router excels in CPU- and network-intensive scenarios but complicates network management. Among overlay plugins Flannel performs best in CPU-intensive tasks, while Cilium is superior for network-intensive tasks. This analysis provides insights for selecting suitable plugin based on workload characteristics.

KW - Cloud networking

KW - Container network interface

KW - CPU usage

KW - Kubernetes

KW - Performance profiling

UR - https://www.scopus.com/pages/publications/105013605116

U2 - 10.7840/kics.2025.50.7.1085

DO - 10.7840/kics.2025.50.7.1085

M3 - Article

AN - SCOPUS:105013605116

SN - 1226-4717

VL - 50

SP - 1085

EP - 1099

JO - Journal of Korean Institute of Communications and Information Sciences

JF - Journal of Korean Institute of Communications and Information Sciences

IS - 7

ER -

Unveiling Kubernetes CNI: An In-Depth Analysis of Networking Performance and Resource Efficiency

Abstract

UN SDGs

Keywords

Access to Document

Other files and links

Fingerprint

Cite this