Year-One Build Log: Everything I'd Do Differently

A year ago this rack was a wire shelf, a surge strip, and a gateway sitting on top of a modem sitting on top of a book. Today it is a 12U wall cabinet with a patch field, a labelled power scheme, and roughly a dozen devices that all have a defined place. In between there were about six weekends of work, four of which I would happily do again and two of which I would like back.

This is the retrospective. Not a build guide — a list of the decisions that turned out to matter, the ones that did not, and the specific points where I paid twice because I guessed instead of measuring. Most of it is not exotic. That is rather the point: the expensive mistakes in a homelab are almost never the interesting ones.

The cabinet: I bought for today's gear, not next year's

I bought a 10-inch cabinet because everything I owned at the time fit in a 10-inch cabinet. That reasoning was sound for about seven months.

The format itself is fine. A 10-inch rack gives you roughly 254 mm of outer rail-to-rail width and something in the 220–235 mm range of actual usable equipment width, depending on how the manufacturer built the rails. For gateways, small switches, mini PCs, and consoles, that is plenty. EIA-310-D still governs the vertical dimension, so 1U is 44.45 mm regardless of whether the rack is 10-inch or 19-inch — the half-width format does not change the height math at all.

What I got wrong was depth, and then height, in that order.

Depth

I bought a 300 mm cabinet. The advertised number describes the frame, not the space between the rail plane and the inside of the rear door. Subtract door hardware and the rail offset and you are realistically working with 240–260 mm of clear space. Then subtract what connectors need: a right-angle C13 wants about 25 mm, a molded RJ45 boot wants 25–30 mm before the cable can begin to turn, and a mini PC with a rear barrel jack and stiff DC lead wants 40 mm or more before the connector stops taking side load.

The rule I now use, having learned it the slow way: take the deepest device you intend to mount, add 50 mm, and treat that as your minimum cabinet depth. If a device is 200 mm deep, you want 250 mm of clear internal space, which means shopping for a cabinet advertised at 350 mm or more. Here is the gotcha: the device that eventually broke my depth budget was not a computer. It was a UPS.

Height

I bought 9U and filled it in five months. Not because I bought more gear than planned — because I badly underestimated the U-count of things that are not equipment. A patch panel is 1U. A blanking panel where the airflow needs interrupting is 1U. A shelf for the one device that will never get a proper mount is 1U. A vented spacer above a warm mini PC is 1U you cannot use for anything else.

By the time the "real" devices were in, roughly a third of my rack units were doing structural or thermal work rather than holding a device. Plan on that ratio. If you need six devices mounted, buy for nine or ten rack units, not six.

Power: the thing I deferred for eleven months

For most of year one my power scheme was a surge strip zip-tied to the rear frame with everything's wall wart hanging off it. It worked. It also meant that every time I moved one device I disturbed four others, and that I had no idea what the rack actually drew.

Two things I would do on day one now.

First, measure. A cheap inline meter told me the whole rack idles around 55 W and peaks near 130 W when the mini PC is compiling and the PoE switch is feeding cameras. That single number would have saved me a lot of guessing. It sets your UPS sizing, and it feeds directly into the thermal math: every watt in is 3.412 BTU/hr of heat out, so 130 W is about 444 BTU/hr that has to leave the cabinet somehow.

Second, buy the UPS before the cabinet, or at least measure it before the cabinet. Consumer line-interactive units in the 1500 VA class are typically rated around 900 W of real load and are frequently 380–430 mm deep. That is deeper than most 10-inch cabinets are internally, which is how I ended up with a UPS on the floor under the rack — the least elegant part of the whole installation and the one I have the least excuse for. If I were doing it again I would either size for a smaller 600–900 VA unit that fits, or accept from the start that power lives outside the cabinet and design the cable entry accordingly.

On PoE, the budget arithmetic is worth doing before you buy the switch rather than after. 802.3af delivers 12.95 W at the device from a 15.4 W port allocation; 802.3at delivers 25.5 W from 30 W; 802.3bt Type 3 delivers 51 W from 60 W. The gap is cable loss, and it is not optional. A switch advertising a 60 W total PoE budget runs four af cameras and nothing else. Mine ran out of budget the week I added a doorbell.

Cooling: I bought fans before I understood paths

My first thermal intervention was to add fans. My second was to remove most of them.

A sealed 10-inch wall cabinet with a solid door has no path. Air warms, rises, hits the top, and sits there. Adding a fan to a cabinet with no intake and no exhaust mostly just stirs the same air and adds noise. What actually dropped my top-of-rack temperature by a meaningful amount was swapping to a vented door and opening the bottom of the cabinet — giving stack effect somewhere to work — and only then adding a small exhaust fan at the top.

The ordering matters: path first, then assistance. If you want a number to sanity-check against, in imperial terms the temperature rise across a moving airstream is roughly ΔT(°F) = watts × 3.412 / (CFM × 1.08). At 130 W and a genuine 20 CFM through the cabinet you are looking at a rise around 20 °F. That is tolerable. At 5 CFM — which is what a fan pushing against a closed cabinet actually delivers — it is not.

The other thing I would change: I would stop mounting warm devices directly above each other. Two mini PCs stacked with no gap means the upper one intakes the lower one's exhaust. A 1U vented gap costs you one rack unit and buys back several degrees. That is a trade I now make without thinking about it.

Mounting: the strategy that held up

This is the part I would mostly repeat.

The decision that aged well was committing early to mounting everything at the rail plane rather than shelving anything. Shelves are quick and they are a trap — a shelved device has no defined position, so it drifts, and every cable behind it has slack that has to go somewhere. Every device I put on a shelf in month two I ended up mounting properly in month eight anyway.

The second decision that aged well: standardizing on the rail hardware before buying mounts. Threaded rails, square-hole rails with cage nuts, and unthreaded rails with clip nuts are three different worlds, and mixing them means keeping three fastener kits and matching them by eye at the worst possible moment. I settled on one and bought a box of the right screws. Cheap, and it removed a recurring small annoyance.

What I would change is sequencing. I mounted devices in the order I acquired them, which meant the patch field ended up in the middle of the rack because that is where there was room when I got around to it. Patch fields want to be at the top or immediately above the switch they serve, because that is what makes patch cable lengths predictable. Mine are not predictable. There is a 1 m cable in there doing a job a 0.25 m cable should be doing, and it will stay there until the next full teardown.

Cabling and labels: cheap then, expensive now

I skipped labels because there were six cables. There are now thirty-one.

The specific failure was not laziness about labelling, it was labelling inconsistently — some cables got a printed flag, some got tape, some got nothing, and the naming scheme changed twice. An inconsistent scheme is worse than none, because you trust it and it is wrong. TIA-606-C exists precisely to stop this, and while the full standard is more structure than a homelab needs, the core idea transfers: pick an identifier format, write it down somewhere that is not your head, and apply it to every single termination including the ones you think are obvious.

The other cabling regret is bend radius. Cat6 wants a bend radius of roughly four times the cable diameter, which for typical solid-core cable lands around 25 mm. Behind a rack with 240 mm of clear depth, a molded boot plus a 25 mm turn plus the device body adds up faster than it looks on paper. I have at least three cables in there that are turning harder than they should because I bought lengths that were slightly too short and routed them straight instead of with a service loop.

What turned out not to matter

Three things I worried about that were not worth the worry.

• Filament choice for printed mounts. I agonized over this. In a cabinet that stays below about 50 °C internally, PETG has been entirely stable for a year with no measurable sag under device weight. ASA and PC matter in hot attics and vehicles. In a conditioned room, they are an answer to a question I did not have.
• Noise, after the first fix. I assumed noise would be an ongoing project. It was one project: replace the two loud fans, add a vented door so the remaining fan does not have to work hard, and stop. Nothing since has moved the needle.
• Aesthetic consistency. I spent real time trying to get every device's front face flush and matched. Then I put the rack in a closet. The devices that are visible are worth aligning; the rest are not.

The short version

If I were starting the same build tomorrow, the changed decisions would be: buy at least 350 mm of cabinet depth and 10–12U of height for a six-device plan; measure total draw and size the UPS in the same week I size the cabinet; establish the airflow path before buying a single fan; put the patch field at the top and work down; and pick a labelling format on day one and never deviate from it.

Everything else — mount hardware, filament, exact device choices — turned out to be reversible. The cabinet, the airflow path, and the naming scheme are the three things that are genuinely painful to change once the rack is loaded, which is exactly why they deserve the up-front thinking.

Wrap-up

Year one cost me about two weekends of avoidable rework, all of it traceable to a measurement I did not take or a plan I made after the fact rather than before it. The gear was never the problem. If there is one habit worth stealing from this list, it is measuring the deepest thing you own before you buy the box you intend to put it in — that single step would have eliminated more than half of what I would do differently.

Building something like this? Browse our most popular mounts — every one is designed, printed, and test-fitted here.