Heat: the fault that breaks nothing but shortens everything

A server that runs too warm does not crash. It slows down and ages faster. Because it reports an error nowhere, it goes unnoticed for months, sometimes years.

Three effects, in order of quietness

Slowing down. Processors reduce their speed to protect themselves. Nothing reports an error; the machine is simply slow. And it is slow exactly when it is working hard, which is when it is least likely to be noticed: "it is under load, of course it is slow", and on we go.

Ageing. Under sustained heat the expected life of every part falls. You do not see this today. You see it two years later, as a failure that came earlier than it should have.

Failing together. This is the one that matters and nobody says it. Heat affects everything inside the box at the same time. Disks bought together, running together and warming together die together.

That is exactly the scenario your disk layout does not protect you from. When one disk dies the repair begins, the repair strains the remaining disks, and those disks have already lived the same length of time at the same temperature.

Where heat comes from in a small server

Usually not from the processor. A closed cabinet, a dusty filter, drives stacked on top of each other with no air gap between them, and the most common one: the machine having been put somewhere with no air.

And the reason is always the same: noise.

The trade nobody plans

It goes like this. The machine is loud, it is annoying. It gets moved into a cupboard or a storeroom. The noise problem is solved.

Six months later the disks are running ten degrees warmer and nobody connects the two events. Because a long time passed between them and no warning appeared in between.

On the day you move the machine, look at whether the place you are moving it to gets air. You will have no reason to look later.

The number to measure is not the peak, it is the sustained value

A brief rise in temperature during a backup is normal. A momentary peak tells you nothing.

What is meaningful is the value it sits at for hours. A system that stays at the same high temperature all day is wearing out far more than one that briefly peaks.

And a practical hint: do not look at a single disk's temperature, look at the difference between disks. If one disk sits consistently warmer than its neighbours, that is not a disk problem but a placement problem. Moving that disk is cheaper than buying a new one.

What Atlas does

Atlas shows each disk's temperature over the last twenty four hours alongside its read and write rates. So you answer not "how many degrees right now" but "where did it sit through the day"; that is exactly the test above.

One design detail deserves saying, because it touches a trap monitoring tools often fall into: these measurements are fetched only when you open that disk's card. For disks sitting closed in the list, no query goes out at all.

The reason is simple and a general principle: a monitoring panel that constantly queries everything turns into a source of load itself. The thing that watches should not be the thing that costs.