While carbon-free energy is crucial to reducing GHG emissions and water consumption, green energy alone is not enough to maintain sustainable operations and address the adverse environmental impact of data centres. Here, Nour Rteil, Lead Application Developer at Techbuyer, discusses the importance of implementing energy efficient servers, for example, to reduce carbon impact on the environment and calls for data centre leaders to improve their energy use/efficiency.
Current state of play
Over recent years we have seen a flux of data centres achieve carbon neutrality by participating in schemes designed to make equivalent reductions of carbon as the ones produced, also known as carbon offsetting. Often, data centres offset their carbon footprint by purchasing renewable energy credits (RECs). Today, data centres are going one step ahead and promising to use carbon-free energy in the future, by relying 100% on renewable energy sources, not only in their on-site operations but also in their overall value chain.
Between the hyperscale data centres, Google was the first major company to reach a carbon neutral state in 2007 and is working towards reaching a 24/7 carbon-free energy target across all its data centres by 2030. Microsoft has also committed to becoming carbon negative by 2030 and is planning to offset all its historical carbon emissions by 2050. Facebook has committed to reaching net-zero carbon across its value chain by 2030 and Amazon has recently pledged to achieve net-zero carbon by 2040.
The limits of the green energy revolution
Shifting to renewable energy has environmental benefits beyond reducing carbon and other GHG emissions. Unlike thermal power plants, wind and solar plants do not require large amounts of water to operate (estimated to be 10% of global water withdrawals by the International Energy Agency). According to Amazon, its renewable energy generation helped to avoid the withdrawal of around 480 billion litres of water in 2020.
Nevertheless, renewable energy relies on natural resources that are not present all the time and are therefore limited in supply. For this reason, data centres need to be mindful of their energy consumption and waste, even if it is generated from a green source. If data centres improve their energy efficiency and consequently reduce their energy consumption, excess green energy could be sold back to the grid and in turn, used to further reduce global emissions.
Optimising overall energy use through servers
Over the past years, data centres have done a great job in improving their infrastructural energy efficiency and this is evident when looking at their reported PUE figures. Google for example reports a 12-month PUE of 1.1 across all their large-scale data centres in all seasons, including all sources of overhead. However, PUE only captures the site’s infrastructural efficiency improvements over time and doesn’t give any indication about the efficiency of IT equipment, which remain the biggest energy consumers in data centres.
Often, IT energy information and utilisation rates remain hidden. The ISO/IEC 30134 series defined two KPIs designated for IT efficiency: IT Equipment Energy Efficiency for servers (ITEEsv) and IT Equipment Utilization for servers (ITEUsv). Although some hyperscalers such as Amazon, Google and Facebook claim that most of their energy efficiency gains are attributed to using more energy-efficient servers (that in some cases are custom-built) and higher utilisation rates, yet, unlike PUE, we don’t see ITEEsv and ITEUsv figures published in their sustainability reports.
This is mainly due to the complexity and time-consuming nature of setting up and measuring the server’s energy efficiency using SERT benchmark as specified by the ISO standard, and having to re-test the server every time it’s reconfigured.
Machine Learning solution
In 2020, Interact was developed to help data centres realise their servers’ energy efficiency without needing to run SERT on every single server on their site. The tool is based on Machine Learning and is trained using published SPECpower results as well as in-house benchmarking results to give power and performance estimates for any server configuration with high accuracy. For a given data centre, the tool provided a server refresh recommendation that resulted in an estimated total energy reduction of 70%. By replacing existing servers with more efficient models, big savings were attained, not only in the total operational energy, but also in associated carbon emissions, cost and rack space. This shows how essential it is for data centres to employ efficient servers that are fit to their needs.
The materials perspective
Besides energy waste, a less talked about concern is material depletion. Data centre buildings and their IT and cooling equipment are material extensive. Historically, data centres have not been vigilant about their materials selection and use. Servers, for example, contain critical raw materials and are designed to last seven to eight years, yet they are refreshed, on average, after only three years. And although there have been considerable advancements in recycling technologies, some server components can’t be recycled so far, and recycled materials are not 100% recoverable.
Therefore, it is necessary at this point to incorporate circular strategies at every business level to reduce the pace of material depletion and the growing e-waste crisis. Google’s Circular Economy encourages maintaining, refurbishing and reusing hardware in its data centres. Facebook is also keen on prioritising sustainable materials and integrating circular thinking into its processes. Unfortunately, many enterprise hardware vendors don’t incorporate circularity into their designs and many enterprise data centres are still hesitant about procuring refurbished equipment, often due to misconceptions about the reliability and performance of refurbished hardware.
While it is important for data centres to run on carbon-free energy and choose suppliers that do too, it is also important to set targets and aim for improving the energy use/efficiency (going beyond PUE) as well as material use and circularity in their operations. For this to happen, additional transparency, accountability and continuous efforts are required from all data centres – especially the hyperscalers.
