There are hardly any situations in our everyday life that take place offline – especially now in the midst of a pandemic. We almost always carry a device with us that is connected to the internet. We rarely store our photos, maps, music or messages on hard drives anymore, but increasingly online. All of this results in huge amounts of data, and while those virtual files are saved in an intangible online cloud, all that cloud data is actually being stored in a physical location too. To try and handle the ever-growing flood of data, internet companies are busy building more and more huge data centres, often covering several tens of thousands of square metres. Google, for example, is currently building its fifth data centre in Europe in Denmark. The buildings are set to house thousands of servers that manage data from Google Search, Gmail and YouTube.
The challenge here is that these data centres consume an extremely large amount of energy, especially when it comes to keeping all that technical equipment from over-heating. To tackle the problem, in 2016 Google commissioned its subsidiary Deepmind to develop a cooling system that is as energy-saving and efficient as possible. Deepmind is a former start-up that was taken over by Google in 2014. It became known a few years ago through its AI program Alpha-Go, which was able to beat the best human players in the world at the strategy game “Go”.
And it seems like Deepmind has been able to achieve impressive results when it comes to developing the cooling system too: According to their own statements, the AI was able to help the US internet giant reduce energy consumption in its huge and complex server buildings by up to 40 percent.
What is the role of AI?
Explained in its essentials, the AI system works like this: The AI solution from Deepmind takes a snapshot of the data centre cooling system every five minutes, composed of information from thousands of sensors. This image is fed into a network that predicts possible scenarios of energy consumption in the near future. The AI system then determines which measures will minimise the energy consumption. These measures are then forwarded as a suggestion to the data centre, where they are verified by a control system and in a final step, implemented.
What is particularly interesting about this system is that the AI works completely autonomously. No human intervention is required. At first, the Deepmind system only developed suggestions for actions that could be taken, which were then manually checked and approved by employees of the data centre. However, it soon became apparent that this method of working was too time-consuming and the human verification step was removed.
Handing over complete control to a computer might sound a little creepy. But Deepmind has various built-in security mechanisms. Every action the AI takes is verified on two levels: first passing its own manually-integrated security requirements and then being sent from the AI cloud to the actual data centre where it must meet the requirements too. Only then can the actions be implemented. It is also possible to switch seamlessly from the AI system to a manual control system at any time.
AI-powered cooling systems – made in China
The Chinese company Huawei has been pursuing a similar approach to Deepmind since 2018. Thanks to their iCooling system, they’ve managed to improve the energy balance of their data centres and reduce power consumption by eight percent. The iCooling system started out by initially analysing a large amount of historical data and identifying the effects of various measures on energy consumption. This enabled a prediction model to be developed. An optimisation algorithm is used to determine the ideal parameters and implement the most energy-efficient adjustments, which significantly increases both speed and effectiveness. The result? Energy savings.
Huawei states that its own cooling system has actually also become safer since the introduction of AI. iCooling is said to be able to predict impending equipment and component failures and thereby warn operating personnel in advance; in the best case scenario, it can even provides suggestions for the best steps to take. According to Huawei, this allows them to detect faults and isolate them extremely quickly, reducing the risk of fires and improving the reliability of the power system.
The two cases presented show how new AI technologies can help to improve energy efficiency. But in order for data centres (and thus digitalisation itself) to become truly sustainable, further changes need to be made. The energy used to run data centres must come exclusively from renewable energy sources and we should look for new ways to use all of the waste heat that’s generated in server farms each day. At RESET, we have already looked at a Dresden start-up that uses waste heat from servers to produce hot water, and even a Paris-based company that uses waste heat from data centres to heat a swimming pool.
And last but not least, it is also up to us users of digital services to take steps to keep our own digital footprints as small as possible. What to know what you can do? Here are our tips on how best to shrink your own digital carbon footprint.
This is a translation of an original article that first appeared on RESET’s German-language site.
This article is part of the RESET Special Feature “Artificial Intelligence – Can Computing Power Save Our Planet?” Explore the rest of our articles in the series right here.
