Nobody sets out to build a seven-node cluster.
You start with one sad little box and a hunch that the cloud is quietly robbing you. That’s the whole trap, honestly. It looks like thrift. It is very much not thrift.
Back at UC Irvine, junior year, too much time between classes, I got curious about infrastructure and figured the fastest way to actually learn the stuff was to run something real instead of just reading about it. So I grabbed a machine, named it tank, and threw Ubuntu Server on it. Why Ubuntu? Because the whole internet swore it was the easy one. (The internet lies. Different post.)
And then I just… watched tutorials. So many tutorials. Late nights, sudo this, systemctl that, breaking the box and fixing the box and slowly getting a feel for how Linux actually holds itself together. Somewhere in there I dumped my paid photo storage for Immich. Then killed a streaming sub and stood up Plex instead. Felt like I’d cracked some code everybody else was too lazy to bother with.
Except. Here’s the bit they leave off the thumbnail: to host your own photos, you need somewhere to actually put them. Drives. And drives aren’t a nine-dollar monthly nibble. They’re one big ugly charge, right now, up front.
So picture it: me, “saving” ten bucks a month by dropping three hundred on hard drives. Genius. By my own math I’d break even around… 2031?
Then a buddy wanted a Minecraft server. And obviously tank was far too important for something that frivolous (it was running, like, two things), so I bought a little AOOSTAR mini PC just to host it. Just for the weekend. You already see where this is going. That “weekend” server never turned off. Neither did the part of my brain that had just discovered you can simply buy another computer.
Because two machines isn’t two machines. Two machines is a cluster waiting to happen.
Trouble was, college won’t let you put down roots. You move every year, and nobody’s hauling a rack up three flights of stairs for a lease that dies in June. So the dream just sat there, half-baked, for a good while.
Then I moved to Oklahoma for work. First place that felt permanent. A corner, an outlet, zero adult supervision. You can guess.
I got a rack. And then, well, since I had the rack, six more nodes. Cheap Dell OptiPlex Micros, the little palm-sized ones that cost about as much as a decent dinner. Boom, seven-node Proxmox cluster. Which meant I needed a real switch. Which meant a patch panel, because the back of the rack had become a nest of furious spaghetti. Which somehow meant a UDM Pro SE, because if I’m doing networking I may as well do networking. The sad little power strip? Retired. Rack-mounted PDU now. Obviously.
Every single purchase made perfect sense in the moment. That’s the con. None of it feels reckless until you zoom out and add it up.
These days it’s honestly kind of legit. Proxmox running Talos and Kubernetes, the whole thing stood up with Terraform so I’m not clicking around by hand, and tank still down at the bottom quietly feeding storage to the cluster and running the media stack that started this whole mess years ago. Looks planned. Was not planned. It was one very long chain of “okay, but now I need…”
The drive tax
Here’s the part that actually stings.
Storage is supposed to get cheaper over time. Mine didn’t. In 2024 I bought two 8TB IronWolf drives, a hundred sixty bucks each. This past April I grabbed two 12TB drives at two hundred eighty-nine apiece. Both before tax. More space, sure, but the price per drive nearly doubled in under two years, which is the exact wrong direction for a thing I was supposedly saving money on.
That’s the secret handshake of self-hosting nobody mentions: the cloud bill you smugly canceled was probably cheaper than the pile of drives you keep buying to avoid it. You didn’t save. You pre-paid. In bulk. At a markup that keeps creeping up.
Total damage? I stopped counting somewhere north of three grand. On purpose. Some numbers are better left unexamined.
Where it’s going
The genuinely fun part is only starting, though. Hosting my own projects on my own metal instead of renting a cloud box every time I want to ship something. Right now each project has its own janky hand-rolled deploy, and yeah, that gets old fast. So the plan is one templated GitLab pipeline. A single boring, repeatable way to push anything. And boring is the whole point. In infrastructure, boring is basically a love language.
So, should you?
Do it. Seriously. No hesitation.
It taught me more real Linux, networking, and Kubernetes than any class ever did. The actual stuff that shows up at work, not the textbook version. Just be honest with yourself walking in: you are not saving money. You’re buying a hobby that happens to hold your photos hostage.
And when you buy? eBay first, Facebook Marketplace right behind it. Half my cluster is secondhand OptiPlex Micros some IT department offloaded for basically nothing. The deals are out there if you’re patient. Your wallet’s still gonna lose. It’ll just lose slower.