The moment everything moved

If you ran VMware in 2023, you remember the feeling. Broadcom closed the acquisition, perpetual licenses disappeared, product lines got consolidated into big subscription bundles, and renewal quotes started arriving that looked like typos. Small and mid-size shops — the ones running a couple of clusters, not a couple of data centers — suddenly found themselves priced like hyperscalers.

The industry response was a genuine migration wave. Proxmox got a flood of attention. Hyper-V got a second look from Windows-heavy shops. And a hypervisor that had been quietly maturing since 2018 found itself with a very large audience: XCP-ng.

I was one of the people who made that evaluation and that move. This blog is the write-up I wish had existed when I started.

What XCP-ng actually is

The short version: XCP-ng is the open-source continuation of the XenServer lineage. Citrix open-sourced XenServer years ago, then began pulling features back behind paid editions in 2017. A group of developers - who became the company Vates — forked it in 2018 as XCP-ng, restoring the removed features and committing the whole platform to open source. It has been developed in the open since.

Under the hood you get:

  • The Xen hypervisor - a true type-1 hypervisor with two decades of production history (it’s what AWS originally ran on).
  • XAPI, the toolstack that makes a group of hosts behave as a pool — one API surface, shared configuration, live migration between members.
  • A management layer, Xen Orchestra (XO), which handles the web UI, backups, replication, self-service, and automation for as many pools as you point it at.

If you’re coming from VMware, the rough mental mapping is: XCP-ng host ≈ ESXi host, pool ≈ cluster, Xen Orchestra ≈ vCenter plus Veeam-style backup, because backup is built into XO rather than being a separate product. That last point is a bigger deal than it sounds - I’ll spend several posts on it.

The current state of play: XCP-ng 8.3 is the LTS release, supported for years and receiving steady update batches, and Xen Orchestra 6 is the current management generation. It’s a stable moment to adopt the stack.

The genuine strengths

  1. The pool concept. Hosts joined to a pool genuinely act as one unit. Configuration, storage attachments, and networking are pool-level objects. Coming from vCenter, it feels less bolted-on.
  2. Backup included. XO’s delta backups, continuous replication, and disaster recovery are part of the platform. For a mid-size shop, deleting an entire backup-vendor line item changes the TCO math dramatically. (And if you want a third-party backup vendor, Veeam now supports XCP-ng natively.)
  3. Open source with a real company behind it. Vates sells support and the XOA appliance, which means the project has revenue and a roadmap — but everything is buildable from source. That’s the insurance policy: no one can do to me what Broadcom did.
  4. The forum. The XCP-ng community forum is one of the healthiest I’ve used — Vates engineers answer threads directly, including the hard ones.

The honest caveats

  1. Ecosystem breadth. VMware’s third-party ecosystem — monitoring vendors, every appliance shipping as an OVA — has no equal. You’ll import OVAs fine, but some vendor tooling assumes ESXi and always will.
  2. Memory management is conservative. No transparent page sharing or aggressive overcommit games. Plan RAM honestly.
  3. Know your storage history. For years the platform’s VHD disk format imposed a hard 2 TiB per-virtual-disk limit, and a decade of forum threads reflects it. Modern QCOW2 support removed that ceiling, but you’ll still encounter the old constraint in older documentation and in environments that haven’t migrated their disks yet — a later post covers this in depth.

Who this stack is for

A good fit: shops running tens to hundreds of VMs per site who want boring, reliable virtualization with integrated backup and a management plane they control. Homelabs, too — the free tier is not crippled.

A worse fit: environments deeply invested in VMware-specific ecosystem features (NSX-style micro-segmentation everywhere, vendor appliances certified only for ESXi).

What this blog covers

Deep dives on the architecture, storage internals, backup design, automation (PowerShell, Ansible, Terraform, Packer against the XO API), and the troubleshooting war stories that only come from running this platform for real. This platform moves fast, so posts note which versions they apply to where it matters.

Next up: what’s actually inside an XCP-ng host — Xen, dom0, XAPI, and friends.