For years my homelab lived on a shelf in the office cupboard. A NUC, a small NAS, a switch with a fan that sounded like a hairdryer trying to leave. It worked, but it grew the way these things do: a second NAS for backups, a box for the hypervisor, another box because I told myself a cluster needed three nodes, and suddenly the cupboard door wouldn't shut and the office was three degrees warmer than the rest of the house.
So I bought a rack. A proper 42U one, second-hand, from a company clearing out after a move. It cost less than a decent monitor and weighed roughly the same as a small car. Getting it down the stairs was the first regret, and not the last.
the garage seemed obvious
The logic was sound on paper. The garage is detached, so noise doesn't matter. It has its own consumer unit, so I could pull a dedicated circuit without arguing with the rest of the house over the kettle. And it is cold, which I naively filed under "free cooling" rather than "a problem I have not thought about yet".
The cold is, in fact, both. Free cooling is real and lovely for about nine months of the year. The trouble is the other three, and specifically the mornings. A garage in a British January sits around four or five degrees overnight. Spinning rust does not love being cold-started at four degrees, and more to the point, warm air carrying moisture meeting cold metal is how you get condensation. The first frosty week I went out to find a faint damp sheen on the side panels and had a small, quiet panic.
The fix was not exotic. I kept the rack running rather than letting it idle cold, because a gently warm rack stays above the dew point, and I added a small low-wattage tubular heater (the kind people put in caravans) on a thermostat set to keep the enclosure above ten degrees. It draws almost nothing and it has earned its place. The other change was discipline: no leaving the garage door open for twenty minutes in the cold while I faff with cables, because that is precisely how you invite a lungful of moist air onto cold aluminium.
power, and the bit i actually got right
The dedicated circuit was the best decision in the whole project. I had an electrician run a 16A radial to a small consumer unit by the rack, with its own RCBO, and fed a rack-mount PDU from it. Everything downstream goes through a UPS, a double-conversion one rather than a line-interactive toy, because the garage is at the far end of the supply and the voltage wanders. I sized it for runtime to shut things down cleanly, not to ride out a long outage. The point of the UPS here is graceful death, not heroics.
Monitoring matters more than I expected. I run a couple of cheap temperature and humidity sensors reporting over MQTT, one at the top of the rack and one at the bottom, plus the UPS over USB. The whole lot lands in a dashboard:
# the sensors I actually watch, paraphrased
- garage_rack_top_temp
- garage_rack_bottom_temp
- garage_ambient_humidity
- ups_input_voltage
- ups_battery_runtime_remaining
The first time the humidity sensor pinged me at 2am I was annoyed. The second time it pinged me it was right, and I had a quiet word with the heater thermostat.
A word on the UPS that I wish someone had said to me first: a line-interactive unit in a garage at the end of a wandering supply will spend its life clicking in and out of boost and trim, and that constant relay chatter wears it out faster than the outages ever would. The double-conversion unit just sits there quietly regenerating clean power, and the difference in my logs is stark. The input voltage graph looks like a seismograph; the output is a flat, boring line, which is exactly what every downstream PSU wants to see.
the regrets, itemised
- Noise carries further than you think. Detached garage or not, on a still night the fans hum across the garden, and my neighbour was too polite to mention it until I asked. I swapped the worst offender (that hairdryer switch) for something with proper temperature-controlled fans and the problem mostly went away.
- Dust. Garages are filthy. Cars bring grit in, the door seal is imperfect, and within a month the filters needed cleaning. I added cheap foam filters to the rack doors and a calendar reminder.
- Walking out there is a tax. It is twelve steps and a cold doorhandle, but psychologically it may as well be a different county. Things I would have fixed in five minutes from my desk now wait until I "go out to the rack", which means they wait longer. Out of sight is genuinely out of mind, and a homelab you don't tinker with quietly rots.
- Spiders. I will say no more, except that a rack with mesh doors in a garage is, to a spider, a luxury apartment block, and I have made my peace with this.
would i do it again
Yes, but differently. The rack itself is a joy. There is a real and slightly embarrassing pleasure in cable management done properly, in everything labelled, in pulling a unit on rails instead of excavating a shelf. The garage was the right place for the noise and the power and wrong for the cold and the distance, and I have spent six months patching over the wrong half.
If you are thinking about it: solve condensation before you solve cooling, run a dedicated circuit if you possibly can, and be honest with yourself about whether you'll actually walk out there in February. The hardware will be fine. It is your willingness to maintain it that the garage quietly erodes.