Server cluster: Proxmox VE with Ceph on four servers

In 2026, a server cluster of four servers with Proxmox VE and Ceph replaced a grown virtualisation environment. Today, more than 150 virtual machines run on it. In the failure tests, the platform absorbed the failure of a server, a switch and a hard disk.

At a glance

Project
Set-up of a highly available virtualisation platform
Period
January to March 2026
Operating status
In operation since March 2026
Purpose
Virtual servers and backup on a common platform
Technology
Proxmox VE, Ceph
Principles
Every network connection duplicated, separate storage network, distributed storage across all servers

Starting point

Virtual servers and backup were to run on a common platform. The previous environment had grown and was to be replaced, and operations were to become independent of VMware. At the same time, no single server and no single switch was to put operations at risk any longer.

Objective

The platform was to cope with the failure of a single server, switch or storage device without operations being permanently interrupted.

Approach

  1. Servers and virtualisation

    Four servers form a cluster with Proxmox VE, on which the virtual machines run.

  2. Distributed storage

    Ceph distributes the storage across all four servers. If a storage device fails, the storage remains available.

  3. Network

    Two switches connect the servers at 25 Gbit. The storage network is separate and runs at 100 Gbit. Every connection is duplicated.

  4. Failure tests

    Before going into operation, a server and a switch were switched off and a hard disk was removed. The virtual machines of the switched-off server started on the remaining servers, operations continued without the switch, and the storage remained available without the hard disk.

Result

With this set-up, you get a platform that absorbs the failure of a server, a switch or a hard disk. If a server fails, the virtual machines restart on the remaining servers. If a switch or a hard disk fails, the systems keep running.

  • more than 150 virtual machines run on the cluster. Period: as of November 2026. Source: virtualisation overview.
  • 3 failure tests passed: server, switch and hard disk. Period: acceptance 2026. Source: measurement during acceptance.
  • 100 Gbit in the separate storage network, 25 Gbit for connecting the servers. Period: 2026. Source: project documentation.

View all project case studies