Running your first home server: what it really costs

My first server was an old tower tucked on the floor, original fan, a hard drive that grated at night. It ran for three years. I switched it off the day I worked out what it cost me in electricity, and realised I had paid several times its purchase price to host a blog that six people read.

Hardware is the most visible cost and the least important one

You always start there, because it’s the fun part. A second-hand mini PC, an ARM board, an old laptop with a dead screen. All of these work. The entry budget is modest, and that’s precisely what makes the project appealing.

The trap is that this figure is the only one people look at. You have to add storage, which should be doubled if the data matters, a UPS if power cuts annoy you, and the replacement of the disk that will eventually fail. A server running around the clock wears its parts around the clock.

I bought twice because I started too small: now I buy the machine that runs what I need, not the tiniest one possible. An undersized device gets paid for in hours of optimisation, and your hours aren’t free even if nobody invoices you for them.

Electricity, a quiet and permanent line

A home server uses little at any given moment, and it does so all the time. That’s the nuance that gets lost at the point of purchase. The difference between a device drawing a few watts at idle and a salvaged tower drawing ten times as much shows up on the annual bill, not on the day’s meter reading.

Two simple rules were enough for me. First, measure before guessing: a plug-in power meter costs the price of two coffees and puts an end to estimates. Second, look at consumption at idle rather than under load, because a personal server spends most of its life doing nothing while it waits for someone to ask something of it.

Electricity prices in France have moved several times in recent years, in a direction that doesn’t suit calculations made in 2019. Redoing the sums from time to time is a healthy exercise, even when it ends up proving the cloud provider right.

Time, the real bill

This is where the arithmetic gets uncomfortable. Installing the system takes an evening. Securing it properly takes a second one. Backups take a third, and then you have to test them, which almost nobody does before the day they’re needed.

Then come the updates, the certificates to renew, the service that doesn’t restart after a power cut, the migration because a major version changed its configuration format. Nothing dramatic on its own. Added together, it looks like unpaid on-call work that you’ve assigned to yourself.

I carry on anyway, and I don’t pretend it’s rational. I host my photos, my notes, my backups. I know where my data lives, which is not a minor luxury when the alternative is entrusting your life history to a company that can change its terms of service on a Tuesday morning. That said, I still pay for online services for other things, because I don’t want to run a mail server.

What I would do differently

Start with one service, the one you really lack. Automate the backup before adding the second. Write the reinstallation procedure while you still have it fresh in your mind, because you won’t have it later.

The old tower is in the garage. It serves as a reminder: the cheapest server is the one you switch off when it no longer does anything useful.