Nour Rteil | Feb. 8, 2021, 1:09 p.m.
Power Usage Effectiveness (PUE) is a widely used metric by data centres as a proxy for energy efficiency. Though PUE could indicate the power and cooling infrastructure overhead compared to IT, it does not provide a reliable overall picture of efficiency as it fails to capture IT efficiency and energy proportionality. This report by Uptime Institute perfectly explains the downside of relying solely on this metric as an efficiency indicator.
Instead of focusing on a single metric, it is important to assess the environmental sustainability of a data center holistically. The Climate Neutral Data Centre Pact in the EU highlights the importance of Governance and these 5 key areas when measuring the sustainability of a data centre:
• Area 1: Energy efficiency, which can be divided into
- Efficiencies in the data centre’s infrastructure (power and cooling)
- Efficiencies in the data centre’s servers
• Area 2: Clean energy
• Area 3: Water efficiency
• Area 4: Circular economy
• Area 5: Circular energy system
• Efficiencies in the data centre’s infrastructure (power and cooling)
Hyperscale data centres are noticeably minimizing the inefficiencies in their power and cooling facilities. Today, Google’s average PUE is 1.11 across all their large-scale data centres. Google was able to achieve this low overhead by following these 5 best practices: continuously measuring PUE, effectively managing the air flow, adjusting thermostat, utilising free cooling, and optimizing power distribution (UPSs and batteries).
• Efficiencies in the data centre’s servers
The results of a study by 451 Research show that AWS’s infrastructure is 3.6 times more energy efficient than the average U.S. enterprise data centre. AWS claims that more than two-thirds of this advantage is attributed to using more energy efficient servers and much higher server utilisation. Google also stated that they consume 50% less energy than average data centres and this is attributed to building their custom servers that produce more operations per Watt. Facebook has also made considerable advancements in server efficiency by using custom built Open Compute Project (OCP) servers.
• Clean energy
Google has a long record of using clean energy. In 2007, Google became carbon neutral and matched 100% of its electricity consumption with renewable energy. Recently, Google made a commitment to operate on Carbon-free energy, 24/7, in all regions, by 2030. Google isn’t the only hyper scale to make pledges in this area, Microsoft has also committed to achieve 100% renewable energy by 2025 and AWS is working on achieving 100% renewable energy for their global infrastructure, as stated in their timeline here.
• Water efficiency
AWS has multiple initiatives to improve their water use efficiency and reduce the use of potable water for cooling data centres. Moreover, Microsoft is committing to be water-positive by 2030. However, there is little information provided by Google on this matter.
• Circular economy
Google has been working on reducing e-waste by following a circular economy approach. By 2017, 18% of Google’s newly deployed servers were remanufactured, and 11% of the components used for machine upgrades were refurbished. In 2017, Google resold over 2.1 million units into the secondary market and diverted 91% of waste from landfills, as reported here. They are committed to achieving zero waste by reducing the amount of waste generated and finding better disposal options. Microsoft has also committed to be zero-waste by 2030.
• Circular energy system
New heat reuse projects and initiatives are being developed by some data centres, where applicable. AWS is supporting Ireland in meeting its 2030 renewable energy targets through the new District Heating Scheme in Tallaght, South Dublin. Facebook also showcased a new heat-recovery system in its Odense, Denmark data centre, as mentioned in their 2019 sustainability report.
Targeting inefficiencies in servers for example, will improve the overall data centre’s energy efficiency tremendously, especially that servers are the main energy consumers in data centres. Data centres are already beginning to target inefficiencies in this area by powering off idle servers, and consolidating servers to increase their utilisation rate, at the expense of reducing availability and redundancy. But there are more ways to target server inefficiencies, like labelling and upgrading inefficient servers that execute low operations per Watt and optimizing the Software that these servers are running and operating on.