How Karmada Powers Multi‑Cloud, Multi‑Cluster Production at Cloud Native Days China 2022

On December 3, the Karmada community held the Cloud Native Days China 2022 multi‑cloud, multi‑cluster production practice session in Nanjing.

Cloud Native Days China, co‑hosted by CNCF and Huawei Cloud, gathered more than 30 enterprises and developers to discuss trends and production practices of multi‑cloud, multi‑cluster environments. Speakers from Huawei Cloud, vivo Internet, Hurricane Engine, China Mobile Cloud, DaoCloud, and Zhejiang University SEL Lab shared their experiences.

Karmada Overview

Karmada, the CNCF’s first multi‑cloud container orchestration project, was open‑sourced in April 2021. It enables users to manage multiple cloud clusters as a single Kubernetes cluster, providing unlimited scalability and simplifying resource pooling.

Speaker Highlights

Huawei Cloud – Ren Hongcai

Presented the community’s vision of continuously enhancing core competitiveness, listening to user feedback, accelerating iteration, strengthening enterprise adoption support, and improving open governance.

vivo Internet – Zhang Rong

Described vivo’s one‑year practice of Karmada, covering cluster operation, continuous multi‑cluster application release, elastic scaling, and gray‑release. The team evaluated Karmada against alternatives using kubemark simulations of 2,000 nodes and selected it for their federated container platform.

Hurricane Engine – Xu Yuanchang

Explained how Karmada is used to improve GPU virtualization, reduce AI training costs, and provide decentralized global consistency and Byzantine‑fault‑tolerant traffic management, enabling rapid multi‑region application delivery.

China Mobile Cloud – Duan Meng & Han Weisen

Shared their multi‑cluster MySQL scheduling solution based on Karmada, using dynamic scheduling and intelligent routing to balance apiserver load, reduce etcd latency, and increase resource pool utilization.

DaoCloud – Zhang Xiao

Introduced the KairShip platform built on Karmada operator, ClusterPedia, and CloudTTY, offering one‑click multi‑cloud cluster creation, cross‑cluster permission management, application distribution, storage provisioning, and fault‑tolerance.

Zhejiang University SEL Lab – Cheng Zhe

Presented the “East‑Data‑West‑Compute” national project’s multi‑cloud collaboration in power systems, using Karmada for deployment strategies, graceful failover, real‑time data sync with Velero, and unified multi‑cluster ingress and service discovery.

Large‑Scale Testing

Karmada recently demonstrated management of 100 clusters, each with 5,000 nodes and 20,000 Pods, confirming stable support for hundred‑fold scale. Huawei Cloud’s Shen Tiecheng shared the test methodology and results, highlighting pull and push modes for efficient large‑scale deployment.

Community Growth

The Karmada community has grown over a year, with more than ten enterprises adopting its features such as cross‑cluster scheduling, resource management, service governance, high‑availability applications, and declarative resource interpretation.

For further information, see the project repository, website, and Slack channel.