AI

Surfing the internet is only possible thanks to data centres. Since the coronavirus pandemic, internet use has skyrocketed. In some countries a staggering download rate of 2.6 terabytes per second was reported after lockdowns were imposed. Researchers and companies are now looking for techniques to make data centres more energy-efficient.

All our online behaviour relies on those data centres, which can be described as large warehouses full of servers on which all kinds of online applications run. The physical layer of the internet consumes an enormous amount of energy. Exactly how much is not entirely clear, but it is expected that by 2030 it could take up as much as 20 per cent of our electricity consumption. The companies operating these facilities are racing against time to make their locations energy-efficient. With this aim, Microsoft recently tested an undersea data centre off the coast of the Scottish Orkney Islands.

Keep it cool

By far the largest data centre in Belgium can be found in Wallonia. At Saint-Ghislain in Hainaut, Google owns a complex that extends over 100 hectares. “We run almost all our applications from here,” says Site Operations Manager Frederic Descamps, “from Google Streetview and Maps to YouTube and of course the search engine itself.”

Much of the energy consumed by this Google data centre goes into cooling the servers. “We are lucky with the mild climate here,” said Descamps. “We can cool the servers using evaporation of water. This is done by letting water run past the servers. The water absorbs their heat and goes to a cooling tower, where it is sprayed into the air. The cooled water flows down the tower, so it can pass the servers again.”

This method turns out to be more efficient than the air cooling that data centres usually employ, which is comparable to an air conditioning system. Efficiency is measured using the so-called PUE ratio, short for power usage effectiveness. This is the ratio between the total energy consumption of a data centre and the energy needed to power the servers. “That ratio must be kept as low as possible,” says Descamps. “You want to use as little energy as possible on top of what is specifically needed for the servers.” In Saint-Ghislain, Google achieves an annual PUE average of 1.08, according to Descamps. That is pretty good efficiency: market research firm Uptime Institute found the average PUE worldwide in 2020 was 1.58.

Bigger is better

The oversized facility in Saint-Ghislain is one of the so-called hyperscale data centres. “These server complexes use more than 50 megawatts of power,” says Marc Decorte of data centre builder Kevlinx, “whilst most data centres in Belgium offer less than 10 megawatts of capacity.”

In Brussels, Kevlinx is working on the development of a new hyperscale location on behalf of the British telecoms company COLT. This future data centre in Brussels will consume up to 68 megawatts – equivalent to the consumption of tens of thousands of households.

And yet, such hyperscale centres are often positive evolutions. Large data centres replace smaller versions or server rooms at the companies themselves. These larger locations are often more efficient, and invest more in cooling technology and newer servers. Kevlinx is now aiming for a PUE of less than 1.2 for this site. In addition, it is investigating how residual heat can be used for heating the surrounding offices, to reduce consumption of fossil fuels even further.

Training AI to save energy

Being energy-efficient is not only about cooling: it is also about the efficiency of the servers themselves and the applications that run on them. For a number of years, many of their functions have been based on artificial intelligence. Applications using AI, like YouTube and Google Translate, require a lot of calculations from computers and therefore consume tonnes of energy.

Usually, an AI system is trained by serving it large amounts of data. It then autonomously discovers patterns and learns to take actions based on them. “An AI system in a self-driving car, for example, gets a lot of video images on which pedestrians are already indicated,” says AI expert Mieke De Ketelaere of imec. “The system learns to recognise them on new images and to avoid them once the car is on the road.”

A new training method eliminates certain calculations if they are not important. This process, called network pruning, saves energy, but how much depends on the AI model and the hardware used during the training process.

Photonics as an energy saver

If investments continue to be made, consumption could remain flat until 2022. However, there are other areas where data centres can save energy. The Dutch company EFFECT Photonics makes another link more efficient: the connection between the data centre and the internet. Data is sent back and forth via fibre-optic cables – ‘travelling’ in the form of optical signals. But data centres themselves only work via electrical signals.

“We design the parts that convert electrical data into optical pulses,” explains Boudewijn Docter, President of EFFECT Photonics. The company enables systems to work more energy-efficiently than before. “With previous systems you had a lot of separate parts, but we now place everything on one chip. This way the transition performs better.”

It means you can get the same amount of bits per second through the fibre-optic cable, but with the advantage of a highly reduced energy consumption. Reducing emissions is after all a major concern. Recent predictions state that the energy consumption of data centres is set to account for 3.2 percent of the total worldwide carbon emissions by 2025 and they could consume no less than a fifth of global electricity. By 2040, storing digital data is set to create 14 percent of the world’s emissions, around the same proportion as the US does today (source: Computerworld). It is here where EFFECT Photonics sees making a significant contribution. Their System on Chip technology provide the desired high-speed, but also the necessary energy efficiency. Their photonic integrated circuits for transceivers enable these mega-datacentres to reduce their impact on society. In this increasingly data-driven economy, datacentres remain one of the most important pieces of our business infrastructure, so they will have to adopt this technology.

Servers go under… water?

Meanwhile, new trends are also emerging. Artificial intelligence, for example, helps to adjust the cooling very precisely according to the temperatures in a data centre. Since 2018, a number of Google data centres have been making full use of AI systems to achieve this.

A data centre must be able to guarantee customers that it will always remain online. To be on the safe side, (environmentally harmful) diesel generators are therefore on standby. Companies are now replacing those generators with hydrogen fuel cells. Microsoft has already conducted tests with this in 2020.

In the future, we may store computer parts or even entire servers in liquids. “Liquid can conduct heat better,” says Marc Decorte. “This ensures that less heat is lost compared to air cooling or water cooling. Such technology is coming in the near future.”

Technological innovations can slow down the acceleration of our energy consumption. In the past few decades our data usage has increased exponentially, but the growth in associated energy consumption has been much slower. Expectations are that data centre energy consumption could remain flat at least until 2022.

“Data centres are getting bigger,” says Ronnie Belmans (EnergyVille, KU Leuven). “Perhaps Google’s site in Saint-Ghislain will consume more energy in the near future than the entire Antwerp site of the chemical company BASF, which is a milestone.”

He adds that there used to be 30 power plants in Belgium, but the trend is toward decentralisation: “Hundreds of thousands of small energy installations will soon be connected to the grid. The solar panels on your roof will provide energy to your neighbours. We can only coordinate all this with a highly developed IT infrastructure. In any case, we need data centres for that.”

The above solutions will help in minimising the energy consumption of data centres, but our internet usage will undoubtedly continue to grow. Advanced cooling, AI, storage in liquids, and photonics will help us tame these big – but very necessary – energy beasts.

Article Source: https://www.eoswetenschap.eu/technologie/hoe-maken-we-datacenters-tegelijk-groter-en-zuiniger