KATECH builds resilient infrastructure for automotive R&D

Korea Automotive Technology Institute (KATECH) runs business systems and automotive software research on a platform built with Proxmox Virtual Environment and Ceph. The hyper-converged infrastructure combines compute and distributed storage with redundancy across the physical network, storage traffic, and cluster communication.

The Korean government and automotive industry established the institute in 1990. Headquartered in Cheonan, Chungcheongnam-do, it operates regional divisions and research centers across Korea.

Its work covers future automotive technologies such as autonomous driving, electric vehicles, hydrogen mobility, and software-defined vehicles (SDVs). The institute also provides component reliability evaluations, testing and certification, technical support, and other services for Korea’s automotive and mobility industries. Reliable IT infrastructure plays an important role in keeping these research, development, and testing activities on schedule.

Resilience from day one

The project started with a clear goal: create a stable virtualization foundation for general business applications, simulation environments, and R&D servers used for vehicle software development, testing, and verification. As a greenfield deployment, the new environment did not replace an existing virtualization platform or require a VM migration.

Because multiple research and testing activities depend on the infrastructure, a failure in the physical network, storage communication, or cluster communication could disrupt important schedules. The new platform therefore needed to minimize single points of failure from the start.

Another requirement was to integrate virtualization and distributed storage into one infrastructure. The solution needed to combine high availability, operational flexibility, and room for future expansion while keeping software costs under control.

Unified compute and storage

After evaluating competing virtualization solutions, the institute selected Proxmox VE with Ceph. The combination enables its IT team to operate compute and distributed storage within one hyper-converged cluster and manage both through an integrated platform.

The financial comparison also favored Proxmox VE. The expected costs for the virtualization platform were approximately 53% lower than for the competing solutions evaluated.

Together with ITZONE, the institute designed an architecture that extends redundancy across the key infrastructure layers. Instead of relying on a single communication path, the design provides alternatives for physical networking, Ceph traffic, and Corosync cluster communication.

Redundancy across every layer

The project ran from September 1 to October 31, 2025. Both teams spent approximately 2 weeks on consulting and architecture design, followed by around 6 weeks of implementation, testing, and system stabilization. Since this was a new-build project, they did not need to migrate existing virtual machines or schedule a separate cutover. The environment entered full production in November 2025.

The platform uses a hyper-converged Proxmox VE and Ceph cluster based on Fujitsu PRIMERGY RX2540 M6 servers. Each node has 2 Intel Xeon Gold processors, 384 GB of RAM, and SATA SSD-based storage. Proxmox VE High Availability helps keep virtual workloads available if a node fails, while Ceph distributes storage across the cluster. The architecture separates the Ceph Public Network from the Ceph Cluster Network and assigns a dedicated VLAN to each.

The Public Network handles client access to storage, while the Cluster Network carries OSD replication and recovery traffic. This separation reduces contention between the 2 traffic types.

The design uses 2 physical switches connected through an Inter-Switch Link. If a network port or an entire switch fails, cluster and storage traffic can continue over an alternative path. Redundant Corosync links provide alternative paths for cluster heartbeat and quorum communication, reducing the risk that a network failure could isolate a node or affect quorum. ITZONE supported the project from architecture design and implementation through testing and stabilization.

Proven in real-world operation

During implementation and stabilization, the teams tested several failure scenarios. After the platform entered production, one of the physical switches failed. The redundant design kept services available, and operations continued without interruption. The cluster has recorded no service-affecting outage since its launch.

The platform now provides a stable foundation for business applications, automotive simulations, and SDV development and verification. Its layered redundancy reduces the risk that a single network component or communication path could interrupt research and testing activities. The integrated Proxmox VE and Ceph architecture also supports future expansion.

Built to grow

The institute plans to expand its use of Proxmox VE as its research activities and cloud service requirements grow. The current architecture allows the IT team to add Proxmox VE nodes and deploy further workloads when needed, providing a flexible foundation for the next phase of automotive and SDV research.

Jun Song, Ph.D.
Senior Researcher, Korea Automotive Technology Institute (KATECH)

About Korea Automotive Technology Institute
KATECH is a specialized automotive research institute established in 1990 through collaboration between the Korean government and the automotive industry. It conducts research into autonomous driving, electric vehicles, hydrogen mobility, software-defined vehicles, and other future mobility technologies. The institute also provides testing, certification, technical support, and component reliability evaluations.

About ITZONE
ITZONE Co., Ltd. is a Proxmox partner providing infrastructure consulting, architecture design, and cluster deployment. For this project, ITZONE handled architecture design, implementation, testing, and stabilization. https://itzone.co.kr/