Data center and UPS

We start with the need, not the drawing: how much power the room needs to handle, how many racks, what redundancy is required, and how much room there needs to be for growth. From there, we lay out the rack plan, routing, power distribution, and cooling so that the parts fit together. It is far cheaper to allocate space for the next rack now than to build the room two years later.

The cooling is included in the contract. The actual work on the cooling system requires authorization and is carried out by our subcontractor, while we are responsible for the coordination and delivery as a whole. We include the heat load from the racks and UPS in the dimensioning, so that the cooling system matches what the room actually emits, and not a loose estimate.

Two independent supply paths, an A-side and a B-side, run physically separate all the way to each rack. Equipment with two power supplies is connected to a PDU on each side, so that a single failed link, a panel or a UPS, does not take down the operation. Redundancy is only real if the two paths do not meet along the way, and that is exactly what the documentation must be able to show.

A PDU is the power strip in the rack that equipment plugs into. There are three levels:

• Basic: distributes electricity without metering. Cheapest, and probably where consumption is known and stable.
• Metering: shows usage per rack or per outlet, so you can see the load before it becomes a problem.
• Managed: can also turn the individual outlet on and off over the network, which is useful when no one is in the house.

We recommend measuring per rack as a minimum. This is the basis for sizing the UPS correctly next time.

A UPS keeps power running when the utility fails or dips. It takes over without interruption and continues to run on battery power until the utility is restored or an emergency generator is started. At the same time, it cleans the power from spikes and fluctuations from the power grid, which over time wear out server equipment and network components.

The type that is called online double conversion or VFI in the data sheets. The system continuously converts the current from the power grid to direct current and back again, so that the equipment always gets a clean, newly generated voltage, regardless of what happens in the grid. In the event of a power failure, the switching time is 0 ms, and the operation does not notice anything. This provides complete decoupling from the voltage fluctuations of the power grid, and therefore it is the topology that we recommend for server rooms and data centers.

We calculate the total power requirement for all connected equipment (kW and kVA) and add an allowance for future growth. The battery bank is then sized based on how long the operation needs to run: minutes enough for a controlled shutdown, or longer if there is no generator. We recommend measuring the actual consumption rather than calculating from the equipment's nameplates, which almost always indicate more than is used.

• VRLA (lead-acid): lower purchase price, but shorter lifespan of typically 3 to 5 years, greater weight and more floor space.
• Lithium-ion: longer lifespan of typically 10 to 15 years, less space and weight, and better tolerance to high room temperatures.

Lithium costs more to buy, but is often cheaper over the entire life of the system because the batteries need to be replaced fewer times.

The system is monitored so that load, battery condition and errors can be followed, and alarms are raised before the operation notices anything. In addition, there is planned service with testing of the battery bank, because a battery that has been untouched for four years does not necessarily last as long as it is designed for. You receive documentation of measurements, tests and replacements, so you can always see when the batteries were last checked.

Yes, and that's a big part of what we do. A system that is sized for yesterday's consumption can be expanded with more UPS modules, a new battery bank or more PDUs without having to rebuild the room. We always start by measuring the actual load, because it is almost always different from the drawing, and then add the expansion on top of what works.