Data centers come in threes, like London buses are supposed to. Or perhaps I should say that data centers ought to come in threes. If you have mission critical software running, clearly it needs a disaster recovery capability.

You could have a whole other data center on stand-by, ready to kick in if a bomb drops on the primary data center. It would have to be a mirror of a kind, but it might not need to be an exact mirror. It could queue transactions up and run them slowly because it doesn’t have a high service level to meet – until it goes into action. There are different strategies you might adopt depending on the software you need to mirror, but the mirror activity need not be as expensive as the primary data center activity.

You could also have both data centers running mission critical applications, so both were primaries and they were mirroring the activities for each other. If one fails, then the other picks up and continues.  That’s fine, but if you get a real disaster that utterly destroys one of these data centers then the other suddenly has no back-up and until you rectify the situation, you are vulnerable. So think in terms of three data centers and you have that possibility covered.

Uninterrupted Power Supply

The model I used in The Death of the Data Center (Part 1) estimated about 20% of the cost being in power distribution and cooling. That’s big, so keeping it low is a good idea. One idea is that, if you have a triple data center configuration, why not do without the uninterrupted power supply (UPS) equipment completely?

You would be vulnerable to power outages, but if your 3 (or more) data centers are truly geographically dispersed, the odds of two losing power at the same time is miniscule. It’s even less if the data center is very close to the power station. The primary cause of power outages is electrical storms, with lightning strikes hitting power lines or trees falling onto power lines. If you minimize the possibilities of that, then maybe you can do without UPS and local generator capability altogether – or perhaps you can reduce it to a very much lower cost.

About 8 percent of power is lost within the data center simply by moving it to the computers. It can be worse. First of all, nearly all the electricity coming in has to be transformed to direct current. Old data centers have older transformers that are less efficient (at least 10 percent less efficient). Old servers have the transformers built into the computer. In a new data center you can put the transformers in a place where you don’t need to cool them. The energy lost in transformation (generally about 10% with very efficient transformers) is taken off the cooling bill. You can run the electric power through the ceilings then some the 8% lost in transmission (which will naturally turn to heat) will not place a load on the cooling system.

Efficient Cooling

The traditional data centers cool the atmosphere of the glass room. Efficient cooling is about focus. You can think of it as cooling the cabinets rather than the whole room and not trying to cool computers that are not turned on. And that means designing the data center accordingly. There’s an optimum temperature which is lower than “the point where the boards bend” but higher than a cooled computer room. Efficient cooling involves finding that temperature and driving the whole cooling system accordingly.

There is really no other conclusion than that the power distribution system and the cooling have to be designed into the fabric of the data center. This is like an oil refinery or a chemical plant. Everything has to be designed with everything else in mind.

See also:

The Death of the Data Center: The Model
The Death of the Data Center: Location, Location, Location
The Death of the Data Center: Power
The Death of the Data Center: Cooling
The Death of the Data Center: Networking
The Death of the Data Center: Server Hardware
The Death of the Data Center: The Software
The Death of the Data Center: Software Optimization