Server Room Commissioning and Acceptance Tests

Installation finishes, everything powers up, screens light and the room is treated as delivered. What has been tested so far, though, is that each device switches on. How the room behaves as a whole, under load and when something goes wrong, is still unknown.

That is what acceptance testing is for. The question it answers is not "does it work" but "what happens when something fails".

Test 1 — Temperature map under load

With equipment running at real load, temperature is measured at several points: bottom, middle and top of the cabinet, the intake and exhaust, and the room itself. Readings taken in an empty room tell you nothing; heat only appears with load.

These figures become the room's reference values. Six months later, "is the room running hotter?" can only be answered against this table.

Test 2 — Power failure scenario

Mains power is cut deliberately, and the following are measured:

• How many minutes the UPS actually holds at the real load — measured on site, not read from a datasheet.
• Whether systems shut down cleanly, in the right order, with databases closing properly.
• What comes back, and in what order, when power returns. Everything starting at once produces an inrush.
• What cooling does during the outage. If it is not on protected power, the room starts heating immediately.

Test 3 — Cooling shutdown test

Cooling is switched off and the time taken to reach a critical temperature is measured. That number tells you how much time you have during a cooling failure. The difference between fifteen minutes and two hours completely changes how you respond to an alarm at night.

The test is kept short; the goal is to see the curve, not to bake the room.

Test 4 — Alarm test

The most skipped and most valuable test. The temperature threshold is lowered temporarily and the alarm is confirmed to fire. Then three questions get answered:

• Who receives the alert?
• Who sees it at two in the morning?
• Is what to do next written down?

An alarm that reaches nobody is the same as no alarm. The same test is run for the water and door sensors.

Test 5 — Labels against drawings

Ten connections are picked at random and their labels compared with the drawings. Where they disagree, the site is right and the drawing is corrected. The check takes five minutes and shortens every future intervention.

Test 6 — Physical checks

• Door lock and access: does the list of people who can get in match reality?
• Cabinet fixings and earthing connections are inspected visually.
• Are blanking panels fitted and cable cut-outs sealed?
• No leftover material, cardboard or packaging in the room — it is both fire load and a dust source.

Readings that are not recorded do not count

The output of acceptance testing is a one-page table: which test, on what date, the value measured, the result. It does two jobs. It documents the room's condition at handover, and it gives you the reference to compare against when something changes six months later.

Open items found during testing go into the same table with a closing date. Anything described as "we will sort that out later" and left unwritten never gets sorted out.

Repeat annually

Rooms drift: equipment is added, load rises, filters clog, batteries age. Repeating the same tests once a year makes that drift visible. The runtime measurement and the temperature map in particular say a great deal when set next to the day-one figures.

The rest of the build

The six stages of a build, backup power architecture and server health monitoring tools.

How Yamanlar Bilişim works

In the rooms we build, acceptance testing is part of handover: measurements are taken, results are entered into the table and the table goes into the handover file. A year later, when the same tests are repeated, the day-one data is there to compare against. Example projects are on the case studies page.

Frequently asked questions