Evolution of Ctrip Cloud Platform: From OpenStack IAAS to Cloud‑Native Kubernetes

Author Bio: Zhou Xinyi, senior R&D manager in Ctrip's Cloud Platform team, leads K8s platform operations, distributed storage, and cloud networking, with over ten years of experience in DevOps and Infrastructure‑as‑Code.

With the widespread adoption of virtualization and cloud computing, underlying IT infrastructure has become increasingly complex. This article outlines Ctrip Cloud Platform's evolution toward cloud‑native architecture, covering design decisions and lessons learned.

Generation 1.0 – OpenStack & IAAS Platform: Ctrip built an IAAS platform on OpenStack to standardize resource delivery. The network relied on OpenStack Neutron and Openvswitch in a traditional three‑tier design. VM provisioning time stabilized at 2‑5 minutes, while end‑to‑end service readiness took 20‑30 minutes due to post‑install steps such as SaltStack agent deployment.

Generation 2.0 – Containerization & Mesos Platform: In late 2015 Ctrip began container trials using novadocker, keeping the same Neutron+Openvswitch network. After validating large‑scale Mesos clusters, Java workloads were migrated, reducing instance creation time to ~30 seconds and achieving >90% containerization of Java services.

Generation 2.5 – Ctrip PaaS + Kubernetes: Growing demands for multi‑language containers, complex scheduling, and auto‑scaling led to a migration from Mesos to Kubernetes in early 2018. Thousands of nodes were deployed, supporting over 10,000 applications across Java, Node.js, Python, and Go, while addressing challenges in Docker daemon stability, kernel issues, and Neutron IPAM bottlenecks.

Generation 3.0 – Cloud Native & Kubernetes: Embracing cloud‑native principles, Ctrip focuses on higher resource utilization (e.g., mixed‑workload scheduling to improve CPU usage), faster provisioning via a container‑image management platform and a new eBPF‑based Cilium network, and stronger application governance through Infrastructure‑as‑Code and Network‑Control‑as‑Code practices.

These initiatives enable centralized resource orchestration, reduce waste, and allow engineers to focus on domain‑specific problems while Kubernetes handles deployment, resilience, and scaling.

Conclusion: Migrating live production systems carries risk and requires meticulous planning, but the open, extensible nature of cloud‑native architectures promises continued gains in efficiency, speed, and control, making it a worthwhile long‑term investment.