When Does Ceph on Proxmox Actually Make Sense?

Ceph is the most tempting button in Proxmox VE. It promises scale-out, self-healing storage without a SAN, and the marketing does not overstate its capabilities. But Ceph is also a distributed system with real operational costs, and choosing it “because it is there” is a mistake we see too often.

What Ceph is genuinely good at

Ceph shines when you need three things at once: horizontal scale, shared storage for live migration, and no dependency on external storage hardware. If you are building a cluster of a half-dozen nodes or more and every node contributes disks, hyper-converged Ceph is a legitimate, proven design.

The costs that are easy to forget

The first cost is network. Ceph traffic should live on a dedicated, preferably 10 GbE-or-faster network, separate from your management and guest traffic. The second is capacity planning: with three replicas you effectively store everything three times. The third is operational maturity — somebody has to understand crush rules, placement groups and recovery behaviour when a disk fails.

When simpler storage wins

For a three-node cluster running a modest number of guests, a well-built ZFS pool with replication, or a small NFS or iSCSI target, is often the more maintainable choice. You get live migration and redundancy without paying the Ceph operational tax.

A decision rule we teach

Ask two questions. Do you expect to grow past four or five nodes? And does the workload actually need scale-out shared storage? If the answer to both is yes, Ceph is worth learning properly. If either is no, start simpler and revisit the decision when the estate has grown.

Ceph is a superb tool. The skill is knowing when not to reach for it.