The mains before the UPS: earth, circuit and heat

The companion article gives electricity a section and settles the right question inside it: a UPS exists to give a machine time to shut down cleanly, not to work through an outage.

We are not reopening that conclusion. What follows is what sits **upstream** of that box, and what the box does not fix: the quality of the voltage, the earth, and the heat that arrives when the power comes back.

Those three share a property that makes them invisible. None of them stops anything on the day it happens. They damage slowly, and the failure shows up months later as a dead power supply, a disk that gives out, or a network port that will not light again.

It is also why this article carries no figure, and section 1 says why in a sentence: the failure is recorded, its cause is recorded nowhere, and nobody can join the two across eighteen months.

The earth connection is the subject where the gap between importance and attention is widest in this whole pillar. It is also the quickest to check.

A socket tester costs a few hundred dinars, plugs into a socket, and lights a combination of indicators telling you whether there is an earth, whether the wires are reversed, or whether something is missing. The test takes three seconds per socket and requires no particular skill.

What a missing earth produces does not look like an electrical problem, and that is what makes it hard to catch. A metal case that tingles slightly when touched. Network ports dying one by one on a switch, for no reason. Unexplained restarts. And, in the case that costs the most, equipment that does not survive a thunderstorm.

The distinction that matters, and it explains why a tester is not always enough: an earth can be present and bad. The wire exists, the light comes on, and the resistance to ground is too high to discharge anything. That shows up only with a measuring instrument, at an electrician’s hands, and is asked for once.

Deal with this point before buying any protection. A UPS, a surge arrester and a "protected" power strip all discharge to earth: without one they are decorative objects, and the surge arrester most completely of all — it has literally nowhere to send what it catches.

The neighbouring article says a battery has a life of a few years and dies without signalling. Here is the routine that turns that sentence into something you will know.

Write the commissioning date on the front of the device in marker pen, the day you install it. That ten-second act is what is missing everywhere: without it the question "since when?" never has an answer and the decision to replace is never taken.

Do not rely on the self-test. Most units run one at start-up and it verifies that the battery exists and produces a voltage — not that it holds a load for eight minutes. A dead battery passes the self-test.

The test that is worth something takes five minutes, once a year, on a Saturday morning: with the usual load connected, unplug the UPS from the wall and see how long it lasts before it starts beeping. Write the figure next to the date. Three minutes in year one, two minutes in year three, forty seconds in year five — that is the curve, and it is visible only if somebody writes it down.

Two details that extend useful life for nothing: a battery hates heat as much as the equipment it protects, so a UPS pushed against an exposed wall or into a closed cupboard degrades twice as fast. And a UPS left unplugged for several weeks — over a holiday, after a move — runs its reserve to the bottom and permanently damages its battery.

A ten-line card, laminated, hung in the equipment room. It is the cheapest deliverable on this page and the one used most often.

On shutdown the order runs from the users towards the data: workstations, then applications, then servers, then storage, then the network, then the UPS itself. The principle is that you never switch off a data store while something is writing to it.

On start-up the order is exactly reversed, and it is the one everybody gets wrong: the network first, then storage, then servers, then workstations. A server powered up before its storage boots believing its data does not exist, and depending on the case it refuses to start or starts in a state that has to be repaired.

Add a waiting instruction between steps — one minute is enough — and a line saying who to call if something does not come back. Those two lines avoid the situation where one person, alone on a Sunday, powers everything up at once and spends the day working out why nothing works.

The card is written once and read aloud the first time it is used. It is the same act as rehearsing a line switchover or a restore: a procedure never executed is a hypothesis, and this one gets executed for real two or three times a year without anybody having to arrange it.

What we refuse first: installing a UPS or a surge arrester on premises whose earth has not been checked. That is not a precaution on principle — those devices discharge to earth, and fitted without one they are sold for a protection they cannot deliver.

We also refuse to size a runtime without a written list of what has to stay on, approved by whoever decides. Without that list the sizing rests on a conversation, and a conversation costs three times too much or not enough, never the right figure.

And we refuse to sell the replacement of a dead device without having looked for what killed it, when two units of the same type have died in the same year. Replacing without looking is comfortable for us: the third replacement arrives by itself.

What we do fits in half a day: the earth test on every socket in the room, the load addition, the shutdown and start-up card written with your people, the dates marked on the UPS and the generator, and section 10’s three lines drafted for your electrician.

And one thing to do this week without us, for a few hundred dinars: buy a socket tester and walk your sockets. You will know in ten minutes whether the protection you have already paid for has anywhere to send what it catches.