Setting Up a Server Room: What Actually Goes In It, and What Drives the Cost

The room already exists in most companies: a small room at the end of the corridor, with a rack, a switch and a server or two, and a split unit on the wall. It runs for years without trouble, until the day the air conditioner cuts out and never comes back, or the power goes off for two hours after everyone has left — and you discover that billing, stock, the cameras and the network were all hanging on that one room.

What follows is what actually goes into fitting out a server room, the order in which the work has to be done, and why two quotations for the same space can differ several times over. You will not find a total figure at the end, because the figure means nothing until the items below have been settled.

1. The rack: depth before height

The first recurring mistake is buying a rack on height alone — "42U" or "27U" — and overlooking depth. Enterprise rack servers such as the HPE ProLiant DL380 Gen11 or the Dell PowerEdge R750 are more than 70 centimetres deep in the chassis alone (around 73 centimetres for the first and 76 for the second, according to each vendor's own figures), and then they need room behind them for cabling and the power distribution unit, and room in front for the door to close. A 60-centimetre-deep wall cabinet — the kind common in network installations — will not take these machines at all. The practical depth for a server cabinet starts at a metre.

The rest of what separates one rack from another:

• Mesh doors, not glass. A server draws air in at the front and pushes it out at the back. A glass door chokes that path, however strong the cooling in the room.
• Blanking panels for empty units. An open U at the front of the rack lets hot air travel back from the rear to the front to be drawn in a second time, so the equipment runs hot while the room itself is cold.
• Spare capacity, planned for. A rack that is full on handover day means the next addition will be a second rack, not another unit.
• Weight and earthing. A rack loaded with servers and UPS batteries can weigh more than half a tonne, and that means something real if the room is on an upper floor. Earthing is not an item to haggle over.

2. Power and the UPS question

The room needs its own feed from the main board, with a dedicated breaker and a cable cross-section suited to the load — not a spur off the floor's lighting circuit. It is done once, and its cost follows the distance between the room and the board more than anything else.

Then comes the UPS, the item guessed at more than any other. Sizing it properly starts from the actual loads, not from the nameplate: a server with two 800-watt power supplies does not draw 1,600 watts — the two supplies share one load for redundancy, not to double it. Add up the real draw of the servers, switches, storage and comms equipment, then leave headroom for growth.

After that, almost everything is settled by a single question:

• Is there a generator in the building? If there is, the job of the UPS is to cover the seconds or minutes of the changeover plus a margin, and the capacity you need becomes relatively modest.
• If there is no generator, its job is to keep the equipment up until it can be shut down cleanly. That means runtime, and runtime means batteries: one or more additional battery cabinets, each with its own price, footprint and weight, and the charger in the unit has to be rated for them. It is these cabinets — not the kVA rating of the UPS — that drive the price in rooms with no generator.

Three points are always forgotten. Do not load the UPS to its full capacity: batteries weaken with age and the load grows. Where mains voltage fluctuates, the right UPS for servers is an Online, double-conversion unit, not the cheaper line-interactive type. And, most important of all, a UPS with no safe shutdown wired to the servers does not prevent the crash — it postpones it until the battery runs flat with the machines still running. The shutdown software usually ships with the unit and works over a USB connection to a single server; covering several machines needs a network card in the UPS, a small extra item next to what it protects, and one that is still left out far too often.

3. Cooling: why an ordinary split unit is the usual mistake

Almost every watt of electricity that goes into a device comes back out of it as heat. So if the room's load is 3 kilowatts, you are cooling the equivalent of about 10,200 British thermal units per hour (one kilowatt equals 3,412 BTU/hr), with the UPS losses, the lighting and the walls on top. That is the basis of the calculation: the equipment load, not the floor area. Sizing cooling by the square metre, the way an office air conditioner is sized, is where the problem starts.

An ordinary split unit is designed for something else entirely, and so it fails in predictable ways:

• It dries the air far more than it should. Equipment heat is entirely sensible heat, whereas a domestic split unit is built to pull moisture out of air that has people in it. The result is very dry air and static electricity.
• Intermittent duty, not continuous. Domestic air-conditioning units are built for hours of operation, not for running all year round. The room needs cooling in January as much as in August, while the outdoor unit may not work well in cold weather without special provision.
• It does not come back after a power cut. Many units do not restart on the same settings when the supply returns, so the room heats up with nobody in it.
• It raises no alarm. The unit stops silently, and nothing tells anyone.
• One unit means zero redundancy. Even routine cleaning means shutting down the only source of cooling.

The alternative is a precision cooling unit, built for continuous running and almost entirely sensible cooling, with filters and alarms. It is a large jump in price and is not justified for every room. The honest middle ground for a small room is two split units with automatic changeover between them, sized to the heat load rather than the floor area, with proper condensate drainage and a temperature sensor that raises an alarm. The mistake is not the split unit itself, but a single unit sized by the square metre, with no alarm and no fallback.

One last point that gets neglected: condensate water. Gravity drainage, or a pump with an alarm, because the wrong route means water above equipment that is running.

4. Cabling and labelling

Cable is the cheapest item in the room and the most expensive one to redo, because it sits inside walls and ceilings. So it is decided once, and decided properly:

• Separate routes for data and power. And where the two routes must cross, let them cross at a right angle rather than run parallel for any distance.
• Patch panel, then patch cord. A wall cable is never terminated straight into the switch. Any later change after that becomes a re-termination instead of swapping a lead.
• The category follows the distance and the plan. Cat6 is enough for most cases, and only Cat6A carries 10 Gigabit over the full length of the link. This decision is about the next ten years, not this year.
• Bend radius and cable-tie tension. A cable bent sharply or tied too tightly gives an intermittent fault two years later, and that is the worst kind of fault to diagnose.
• Test every link and hand over the results, with every cable labelled at both ends and a written schedule handed to the client.

Labelling is the difference between a fault cleared in ten minutes and one that eats half a day. On a job our network teams carried out across the Al-Raya Market branches, terminating the runs onto the patch panel and labelling them at both ends was part of the handover itself, not a cosmetic extra.

5. Access and monitoring

A solid door and a good lock are the minimum. Beyond that, the question is who entered the room and when. A card or fingerprint access control system leaves a log that can be reviewed, and a camera on the door completes the picture.

But the most important part, and the cheapest, is environmental monitoring: a temperature and humidity sensor that sends an alert, a water leak sensor near the air-conditioning unit, an alert when the door opens, and monitoring of the UPS and its batteries through that same network card. The fault that costs more than any other is the silent one — an air conditioner that stops on Thursday evening, with nobody any the wiser until Saturday morning.

And a server room is not a storeroom: cardboard boxes and old chairs are a fire load and a source of the dust that blocks equipment filters.

6. Fire

Detection first: smoke and heat detectors tied into the building's alarm panel, and above the false ceiling and under the raised floor where those exist, because a fire in a room like this usually starts in a cable or a power supply before it shows in the open space of the room.

Suppression, though, is a cost tier of its own. A clean agent system — FM-200, or FK-5-1-12, known as Novec 1230 and available from more than one manufacturer, or the inert gases — needs a room that genuinely seals, a pressure relief vent, and an interlock that shuts the air conditioning down and closes the air openings before discharge. A gas system in a room that breathes from every direction is money paid for no result. The sensible alternative for a small room is good detection tied into the building alarm, with a hand-held clean agent extinguisher, provided the decision is consistent with the Civil Protection requirements for the building — a body to be consulted before the design, not after it.

7. The order of works

The most expensive mistakes in this work are not mistakes of choice but mistakes of sequence. The order that saves real money:

• First: the survey and the drawing. A site visit, an inventory of the equipment and its loads, the growth plan, then a room layout, a bill of materials and a programme — before anything is bought.
• Second: the builder's work. The door, closing off openings that are not needed, finishing the floor, clearing out everything that is not IT equipment.
• Third: power. The feed, the board, the earthing and the outlets, before any equipment is mounted.
• Fourth: air conditioning. Unit positions, pipe routes and drainage. These routes cross the room, so they cannot come after the racks are in place.
• Fifth: cable trays, pulling, termination and testing.
• Sixth: the racks, the power distribution units and the patch panels.
• Seventh: installing the equipment, connecting it and labelling it.
• Eighth: access control, detection and sensors.
• Ninth: testing, the labelling schedule, the as-built drawings and handover.

Every time that order is broken, the bill is paid twice: breaking into a wall after the floor has been finished, moving a rack to run a drain pipe, pulling a cable after the ceiling has been closed. And on a site that is already operating, working in phases and outside business hours becomes a real item in the cost, not a detail.

What makes one room cost several times another

Two quotations for the same space can differ several times over, and the difference is almost entirely in this list:

• The level of redundancy. The biggest multiplier of the lot: one cooling unit and one power path, against backup cooling and a dual feed.
• Whether there is a generator or not, because that decides the capacity of the UPS and the number of battery cabinets it needs.
• The type of cooling: a single split unit, two with changeover, or a precision cooling unit.
• Fire: detection only, or gas suppression with room sealing, venting and interlocks.
• The state of the room before you start: a room that is ready, or building partitions, a raised floor and a new door.
• Distances: how far away the main board is, the lengths of the cable routes, and the number of points.
• The number of racks and their quality, and the power distribution units inside them.
• A site that runs and cannot stop, so the work goes ahead in phases and outside normal hours.
• What is handed over at the end: test results, a labelling schedule and drawings, or simply installed equipment.

Which is why asking for "the price of a server room" before these items have been settled gives you a figure that cannot be compared with any other figure.

Where to start

Start by taking stock of what you actually have and what you expect to add over the next three years, because that inventory alone decides the rack, the UPS and the cooling together. Then ask for a drawing and a list of items, not a total.

At ZEMAVO we begin every case with a site visit: measuring the space, listing the loads, checking the power source and the point where the internet line enters the building, then a drawing, a bill of materials and an agreed order of works. You can read the details of our server room and data centre fit-out service, or of server supply and installation if the machines themselves are part of the decision. And for anyone comparing quotations right now, our article on server prices in Egypt takes apart the items that make the difference between one quotation and the next.

If you already have a room and want to know what it is missing before you spend anything, get in touch to arrange a visit and a survey.