How to Build an Energy Efficient Data Centre

How to Build an Energy Efficient Data Centre

Data centres are among the most prolific consumers of energy in the world. And with 5G and edge computing now on the horizon, data centre power needs are only going to increase. The question before us is how to build more energy efficient data centres so that we do not cripple ourselves.

Written by Anthony Carter, MD, Connotations Publishing Ltd. Published Tuesday, 10 November 2020 10:36

Data centres are among the most prolific consumers of energy in the world. And with 5G and edge computing now on the horizon, data centre power needs are only going to increase. The question before us is how to build more energy efficient data centres so that we do not cripple ourselves.

The key to figuring this out is focusing on something known as power usage effectiveness (PUE). A data centre's PUE ratio reflects how much of the total energy consumed goes to computational loads as compared to everything else. Computational load is the actual work done by servers. Everything else includes lighting, cooling, and so forth.

Improving PUE allows for more computational work without a commensurate increase in energy consumption. If we do not improve this ratio across the board, the amount of energy our data centres consume could eventually get out of control.

Size Does Matter

In a perfect world, the ideal PUE is 1.0. This is to say that 100% of the power consumed goes to the computational load. Whether or not it is achievable is a matter of debate. It seems highly unlikely given the fact that there is always some loss of energy when operating machines.

Having said that, shooting for 1.0 is a worthwhile goal. Getting closer means more energy going to computational load and less energy to everything else. Believe it or not, here is where size really does matter. Larger facilities – sometimes known as hyperscale spaces – tend to have lower PUEs.

A 2016 study from Lawrence Berkeley National Laboratory (USA) showed that a small IT room had the highest PUE in excess of 2.0. On the other hand, a hyperscale facility tended to be closer to 1.2. The study identified four other facility size between the two extremes.

In short, larger data centres tend to have lower PUE ratios. Why? There could be any number of reasons, but the most likely culprit is cooling. That same study demonstrated that the smallest of facilities dedicated up to 50% of their energy consumption to cooling while the largest facilities were closer to 13%.

Larger Facilities Over Smaller Ones

The study suggests that one of the primary methods for increasing data centre efficiency is to build larger facilities rather than more smaller ones to house the same number of servers. In other words, take advantage of the economics of scale. Ten smaller data centres will consume more energy for cooling than a single hyperscale centre housing the same number of servers.

Adopt Artificial Intelligence Solutions

Humanity has discovered over the years that computer systems tend to be far more efficient than human beings. In light of that, we are starting to see a push for completely autonomous data centres where power, cooling, and energy efficiency are concerned. Google built one such facility back in 2016.

The data centre relies on artificial intelligence (AI) to maximise energy efficiency. It continually monitors energy consumption in every part of the facility – and adjusts accordingly. The key is predictive analytics.

Google's system takes a snapshot of the cooling system every five minutes utilising data from thousands of sensors. That data is then fed into a neural network that analyses it and predicts future energy consumption. The system can then automatically modify controls to reduce energy consumption where possible.

Consider More Sustainable Solutions 

Sustainability is a concept that looks to preserve natural resources by finding better ways to do things. It is a concept with multiple applications for building more efficient data centres. A good example is how we go about cooling new data centres after we build them.

More than ever before, modern designs call for using some form of natural air cooling. That is to say that air is drawn in from the outside, circulated through the facility, and forced back out along with warm air. This model works best in cooler environments. So instead of building new data centres in warmer and more humid locations, building in cooler and drier locations makes cooling easier and more efficient.

The opposite side of that equation is harnessing the heat that data centres do produce. There is no need to allow waste heat to escape into the air. It can be captured and used for things like heating the facility or supplementing municipal heating. It could even be captured and used to turn an electricity-generating turbine.

Install the Latest Hardware

Finally, building efficient data centres requires installing the latest hardware. This is not limited just to servers, either. It includes efficient heating and cooling systems, lighting systems, fire suppression systems, monitoring systems, and so forth. Everything that consumes power needs to be up to date for energy efficiency.

We can build more efficient data centres, and doing so is not a pipe dream. Actually, it is a necessity. If we don't get our data centre power consumption under control, our technology may evolve faster than our ability to power it.