Two years ago I put a proper 42U rack in the garage. Not a shelf of mini PCs, not a cupboard with a switch zip-tied to the back of it, an actual rack on castors with rails and a cable manager and ambitions. This is the honest retrospective, the one with the regrets in it, because the glossy homelab posts skip the part where you're standing in an unheated garage in December cursing a cold-aisle that doesn't exist.
I don't regret having a homelab. I regret several specific decisions about this homelab, and they're the useful part, so let's do those.
the garage is a thermal liability, in both directions
I chose the garage for the obvious reasons: it's away from the living space, the noise doesn't matter out there, and there was room. All true. What I didn't think hard enough about was temperature, and a garage is bad at temperature in both directions.
In summer it bakes. With the door shut on a hot day the ambient out there climbed well above what I was comfortable seeing on the intake sensors, and the kit responded by spinning fans to a roar and, on the worst afternoon, thermally throttling. In winter it's the opposite problem, and the opposite problem is sneakier. Cold air holds less moisture, but a garage that swings from cold nights to warmer days is a condensation machine, and condensation on a powered board is how you get a fault that's impossible to reproduce because it only happens at 6am in January.
I now run a cheap temperature and humidity sensor in the rack that feeds into my monitoring, and I have alerts on both ends of the range. If I were doing it again I'd have planned the airflow before I planned the rack. The order I actually did it in, buy rack, fill rack, then worry about heat, is exactly backwards.
noise I was right about, power I was wrong about
The noise call was correct. Enterprise gear in 1U and 2U chassis is genuinely loud, the small high-RPM fans produce a whine that's fine in a datacentre and intolerable in a house, and putting it in the garage meant I never had to care. That decision has aged well. If anything it's the one thing I'd repeat without hesitation.
Power is where I was naive. I costed the hardware and completely failed to cost the electricity, and a rack that idles at a few hundred watts adds up to a number on the bill that I'd rather not type out here. I've since done the obvious things: consolidated workloads so fewer physical boxes run, swapped a couple of power-hungry old servers for lower-wattage kit that does the same job, and put the whole rack behind a metered PDU so I can actually see what each thing costs me rather than guessing. The single biggest saving was admitting that two of the servers existed mostly to justify the rack, and turning them off.
A short list of what actually moved the power number, in order of effectiveness:
- Turning off machines I'd convinced myself I needed and demonstrably didn't.
- Replacing 2010s-era enterprise servers with modern low-power hardware. The performance-per-watt difference over a decade is enormous.
- Setting CPU governors sensibly so things idle low instead of holding a high clock for no reason.
- Measuring, via the metered PDU, so the decisions were based on watts rather than vibes.
cabling: do it once, properly, or do it forever
Here's the one I'm proudest of and the one that took the most discipline. I ran the cabling properly. Patch panels at the top, structured runs down the back, every cable the right length and labelled at both ends, colour-coded by purpose. It was tedious and it took a full weekend that I'd rather have spent doing literally anything else.
It has paid for itself ten times over. Every change since has been a five-minute job instead of an archaeological dig. The version of this hobby where the back of the rack is a bird's nest of mismatched patch leads is the version where you eventually unplug the wrong thing and take down your own DNS at the worst possible moment. Ask me how I know, on a different rack, years ago.
The labelling specifically is the bit people skip and shouldn't. A label printer is cheap. Future-you, tracing a fault at midnight, will weep with gratitude that present-you spent twenty quid and an afternoon.
the regret I didn't see coming: scope
The biggest regret isn't thermal or electrical. It's that a 42U rack is an invitation. Empty U's are a vacuum that demands to be filled, and a rack with space in it is a standing argument to buy more things to put in it. A smaller rack would have imposed a discipline that the big one actively undermines. I bought 42U because it was barely more expensive than 24U, and I have spent two years finding reasons to fill the difference.
If I started again I'd buy a rack slightly too small for my plans, not slightly too big. Constraint is a feature. The homelab that fits in a 12U wall-mount cabinet is, for most people, the homelab they actually need, and it doesn't whisper "you have room for one more server" every time you walk past it.
would I do it again
Yes, but smaller, cooler, and with the electricity bill costed before the first purchase order. The rack has taught me more about real operations, airflow, power, capacity planning, the genuine cost of running things, than any amount of cloud where someone else absorbs all of that and hands you a tidy invoice. There's a particular kind of learning that only happens when the consequences are physical and in your garage.
I just wish I'd spent the first weekend planning the heat and the power, and the last two years resisting the empty U's. The cabling I got right. The everything-else I'm still adjusting. That's homelabbing: it's never finished, and the regrets are where the actual lessons live.