For a while I was doing what most people do — relying on a combination of cloud storage and hoping for the best. Not a real backup strategy, just a vague feeling that things were probably fine. When you’re running a homelab with VMs, containers, and services you actually depend on day to day, that stops being acceptable pretty quickly.

I eventually got around to building a proper backup setup using Proxmox Backup Server running on a small dedicated machine. It took a weekend to get right but now it just runs quietly in the background doing its thing. This is what I built, why I built it this way, and what I’d do differently if I were starting over.

The Hardware

I didn’t want to spend a lot on the backup server itself. It runs 24/7 but it doesn’t need to be fast — it just needs to be reliable, power efficient, and have enough storage to actually hold backups. A Dell OptiPlex 3060 Micro hit all of those marks.

These small form factor business machines are underrated for homelab use. They’re quiet enough to sit in a living room, they sip power compared to a full tower, and the i5-8500T specifically is a solid low-wattage chip that handles the compression and deduplication workload without breaking a sweat. I picked one up used for well under $200 and added the drives separately.

The 4TB spinning disk is where the actual backups live. HDDs are still the cheapest option per gigabyte for this kind of workload — the data sits there, gets written occasionally, and doesn’t need fast random reads. It does the job.

Why Proxmox Backup Server

The short answer is that my entire homelab runs on Proxmox VE already, so PBS was the obvious choice. The integration is native — you add it as a storage target in your Proxmox cluster and from that point backup jobs are just a few clicks to configure. No third-party agents, no weird compatibility issues.

The longer answer is that PBS does a few things really well that matter for this kind of setup. Incremental backups mean after the first full backup, subsequent runs only send what changed. That keeps backup windows short and storage usage reasonable. Deduplication on top of that means if you have ten VMs sharing similar base OS files, you’re not storing ten copies of the same data — just one copy with references.

Restores are fast. That part matters more than people realize until they actually need to restore something. Pulling a VM back from a local PBS instance takes minutes. Pulling it from a cloud backup can take hours depending on your connection and how much data is involved.

Storage Layout

I kept the storage layout simple on purpose. The OS runs on the 500GB SATA SSD — more than enough for the PBS installation and its metadata. The 1TB NVMe handles the chunk store index and any temporary operations that benefit from faster IO. The 4TB HDD is the actual backup repository where all the data lives.

Could I have consolidated some of this? Probably. But separating OS from data from performance layer makes the whole thing easier to reason about and easier to expand later. If I fill up the 4TB I just add another drive to the repository. No reconfiguring anything else.

Why I Don’t Use Cloud Backups

I’ve used cloud backup services before and they work fine for certain things. But for a homelab running multiple VMs they stopped making sense for me pretty quickly.

The Cost Compounds

Cloud storage pricing looks reasonable until you actually start storing meaningful amounts of data. Then it compounds. You’re paying monthly forever, the bill grows as your data grows, and if you ever want to get a large restore you’re either waiting a very long time or paying egress fees on top of the storage costs. The upfront cost of a small dedicated backup machine with a 4TB drive pays for itself in a year or two compared to most cloud options.

Restore Speed Is Not a Joke

If something goes wrong in my homelab at midnight and I need to restore a VM, I want it back in ten minutes, not two hours. Local restores from PBS are genuinely fast. That speed difference is the thing cloud backup advocates tend to gloss over when comparing options.

Everything Stays Local

All my data stays on hardware I own, in my house, under my control. No third party can access it, no service can go down and take my backups with it, no policy change can affect my retention. That matters to me more than the convenience of cloud access from anywhere.

Power and Noise

This machine runs around the clock so I cared about power consumption. The i5-8500T is rated at 35W TDP and in practice the whole system pulls somewhere around 15-25W at idle, which is what it’s doing most of the time. Over a full year that’s a pretty small addition to the electricity bill.

Noise is basically nothing. It sits in the same room as my main server and I genuinely forget it’s there most of the time. Small form factor machines like this are well suited to any space where a full rack or tower would be disruptive.

Final Thoughts

I put off doing this properly for longer than I should have. It felt like a project that required a lot of planning and the right moment to sit down and figure it out. In reality it took a weekend, most of which was waiting for the initial backup jobs to complete.

If you’re running a homelab and your backup strategy is “it’s probably fine” — it’s worth spending a weekend on this. Losing a VM you’ve spent hours configuring because you didn’t have a backup is a uniquely frustrating experience. Setting this up once means you don’t have to think about it again.

There’s also something genuinely satisfying about watching a backup job complete and knowing exactly where that data lives, who has access to it, and how fast you could get it back if you needed to. That kind of certainty is hard to put a price on.