Your own server with Proxmox: what to virtualise and what not to
Standing up your own server to save on a cloud bill is a bad reason. Add up electricity, internet, hardware and your own time, and the savings usually vanish.
The good reason is different: understanding how things work. In the cloud you press a button and a service appears. On your own server the network, the disks, backups and updates are all your decisions. That knowledge pays off later in any environment.
I have been running Proxmox for a few years. This article is what I took from it.
What Proxmox gives you
In short: running several isolated systems on one physical machine and managing them from a browser.
The main value is not the control panel, it is two other things.
The first is snapshots. Before a change you save the state and roll back with one click if it goes badly. That drops the cost of experimenting sharply.
The second is centralised backups. Instead of a separate script inside every machine, everything is backed up on one schedule.
VM or container: the deciding criterion
Proxmox supports two kinds of "guest", and this is where newcomers usually get confused.
An LXC container is the lightweight option. It shares the host kernel, starts fast and uses little memory.
A virtual machine is full isolation. Its own kernel, its own drivers, more resources.
My criterion is simple: container by default, VM when there is a specific reason.
Cases where you need a VM:
- You need a different operating system
- Something complex runs inside Docker — containers inside containers work, but need configuration and make troubleshooting harder
- You need to pass a device through directly
- The system needs kernel-level tuning
Everything else is a container. Nginx, a small database, monitoring, a bot, a static site. In a container they come up in seconds and take little space.
A practical detail: create containers in unprivileged mode. That is the default, and changing it needs an explicit reason. A privileged container sits closer to the host and its security boundary is weaker.
Networking is the part that eats the most time
At first everything is simple: guests reach the home network through a bridge and each gets an address.
That works and is fine to start with. But two things are worth settling early.
The first is that addresses must be stable. If an address changes on every reboot, all your configuration breaks. Give each guest a fixed address or pin it on the router.
The second is that external access should go through one point. Opening a separate port per service becomes unmanageable quickly. I put a reverse proxy in front: it handles certificates and keeps routing in one place.
If you need isolation between services — a guest network or a test environment, say — Proxmox supports VLANs. But do not start there: work with a plain bridge first and segment when the need appears.
Backups: the only correct definition
This is the most important section in the article.
A backup is not a copy. A backup is a copy you have restored.
Configuring backups in Proxmox is easy: set a schedule, pick a storage target, done. And that ease is exactly what makes it dangerous, because it gives you confidence — untested confidence.
Three rules I consider mandatory.
1. The copy must live on different physical hardware. A backup on the same disk protects you from your own mistakes, not from a disk failure.
2. Test the restore. At least quarterly, restore one backup into a new guest and check that it comes up. That is an hour of work, and it tells you the backup actually works.
3. Know what is not being backed up. Some data lives outside the guest — on an external disk or a network share. It does not enter the automatic backup, and that is a common hidden gap.
An addition: a snapshot is not a backup. It lives on the same disk and disappears with it. Snapshots are for experiments, not for recovery.
What not to keep on your own server
This matters too, because the temptation to bring everything home is strong.
Important data in a single copy. Family photos, documents, work projects — a copy must exist elsewhere as well. A home server is a convenience, not a reliability guarantee.
A service that must run continuously. At home the power goes out, the internet drops, you upgrade hardware. If someone's work suffers when it stops, put it in the cloud.
Internet-facing data that is not yours. Storing customer data on a home server is a question of liability, not technology.
In my own setup, production workloads and my personal experimental environment are clearly separated. That separation is not for convenience but for safety: a mistake on the machine you are experimenting with must not touch a running service.
What I learned over time
Three things.
Write documentation. Six months from now you will not remember what runs where. I keep a plain text file: each guest, its purpose, its address, and why it exists. That file is more useful than any monitoring dashboard.
Put updates on a schedule. "The server is running" does not mean "the server is safe". Make monthly updates a habit — snapshot, update, verify.
Fewer guests is better. At first, creating a separate container per service looks appealing. Then each one needs updating, backing up and watching. Keeping related services together is often the better call.
A final thought: your own server is primarily a learning tool. It teaches you things about networking, disks, backups and isolation that the cloud never will. And that knowledge is exactly what pays off most later, in production.