Power Usage Effectiveness (PUE) has long been the standard metric for evaluating the energy efficiency of data centers. Introduced by The Green Grid, PUE is defined as the ratio of total facility energy consumed to the energy consumed by the IT equipment. While PUE has been instrumental in driving awareness and improvements in data center energy efficiency, it is not without its shortcomings.
In 2013, ASHRAE and The Green Grid published the article PUE: A Comprehensive Examination of the Metric to highlight the basis and theory on how it is to be applied. Although is has a lot of information on the theory and application, it still leaves a lot to be desired on consistency and how to apply for more beneficial energy efficiency. It also concentrates on existing facilities, where measurements of PUE are meant to be self-introspective only; it simply chose to ignore how to level PUE for new facilities and comparisons to others. Additionally, partial PUE and other definitions that were an attempt to solve for comparisons seem to be rife with misapplication.
Understanding PUE
PUE = Total Facility Energy / IT Equipment Energy
Therefore, a lower PUE value indicates greater energy efficiency, as it signifies that a larger proportion of the energy is being used directly for computing rather than for cooling, lighting, or other ancillary functions. The best value that data centers should strive to reach is 1.00; however, this means that 100% of the energy for the facility is being used by the IT equipment, a near-impossible feat.
The Limitations of PUE
1. Incomplete Representation of Efficiency: PUE focuses solely on the energy consumed by the data center's infrastructure versus the IT equipment, neglecting other critical aspects of overall efficiency. It does not account for the actual performance or productivity of the IT equipment itself. A data center with a low PUE might still be inefficient if its servers and other IT hardware are not optimized to reduce energy consumption.
2. Lack of Consideration for Renewable Energy: PUE does not differentiate between energy sources. A data center powered entirely by renewable energy (e.g., solar, wind) or powered by fossil fuels have the same PUE. This limitation overlooks the environmental impact and sustainability efforts of data centers that invest in green energy solutions.
3. Variation with Load Changes: PUE can vary significantly with changes in data center load. During periods of low IT load, the energy consumption of supporting infrastructure (cooling, lighting, etc.) remains relatively constant, which can result in a higher PUE. Conversely, during high IT load periods, the PUE may appear more favorable. This variability can make it difficult to compare PUE values across different data centers or even within the same data center over time.
4. Neglect of Other Resource Utilization: PUE does not consider the utilization of other critical resources such as water and space. Water Usage Effectiveness (WUE) and Space Utilization Effectiveness (SUE) are additional metrics that provide a more comprehensive view of a data center’s efficiency. For example, a data center might have a low PUE but high water usage for cooling, indicating tradeoffs that PUE alone would not reveal.
5. Encouragement of Suboptimal Practices: Focusing exclusively on lowering PUE can lead to suboptimal practices. For instance, operators might operate their data centers to achieve a better PUE score even though total energy and cost may increase, which is counterproductive if the energy exceeds what is necessary for optimal operation. This approach can also lead to an underestimation of the potential for efficiency improvements in IT equipment itself.
6. Not Accounting for IT Equipment Age and Efficiency: PUE does not account for the age or efficiency of IT equipment. Older servers and hardware tend to be less energy-efficient compared to modern, optimized equipment. A data center with state-of-the-art IT hardware might have a higher PUE compared to one with outdated equipment, despite being more energy-efficient overall.
7. Geographical and Climatic Variations: PUE is influenced by the geographical location and climate in which a data center operates. Data centers in cooler climates may have an inherent advantage due to lower cooling requirements, whereas those in warmer climates might have higher cooling demands, affecting their PUE despite potentially efficient operations.
8. Age, capacity loading, size of facility, infrastructure redundancy and resilience, and more are also not considered for PUE measurements. Density and configuration of the design are also important factors ignored by the metric.
Towards More Comprehensive Metrics
Given the limitations of PUE, there is a growing recognition of the need for more comprehensive metrics to evaluate data center efficiency. Some of these include:
1. Energy Reuse Effectiveness (ERE): ERE measures how effectively a data center reuses its waste energy. Unlike PUE, ERE accounts for energy that is repurposed, such as heat recovery for heating nearby buildings.
2. Carbon Usage Effectiveness (CUE): CUE measures the total carbon emissions produced by the data center relative to the IT equipment energy consumption. This metric provides insight into the environmental impact of the energy source.
3. Water Usage Effectiveness (WUE): WUE evaluates the amount of water used by a data center relative to the IT equipment energy consumption. This metric is crucial for understanding the sustainability of cooling methods that rely heavily on water.
4. IT Equipment Utilization Metrics: Metrics such as Compute Power Efficiency (CPE) focus on the performance and efficiency of the IT equipment itself. By evaluating how effectively IT resources are used, these metrics provide a more complete picture of data center efficiency.
PUE is meant to provide a baseline of measurements with which to make improvements for an existing facility to itself, but more and more it has been used to compare against other facilities when most in the industry realize that talking about PUE performance is relative and should be ignored unless comparing the same facility (and same measurements) from a past to present change.
The measurements for PUE are well defined for where they are to be taken and with different levels of granularity. Note also that more granularity almost invariably means that the PUE will be greater, therefore many would rather opt out of the expense and time to obtain such detailed measurements if the result will be marginally worse.
To use PUE as a design tool also makes sense, as this is often the baseline to understand the IT power, peak power, and average operational power planned for the facility. This worst case scenario works for the overall loads as well as when aggregating smaller blocks into a larger total load for a data hall, building, or site.
In order to better represent PUE – whether partial, annual, peak, or other – more detailed standards and calculations should be developed. These methods to should be able to isolate component and system efficiencies with the ability to ignore or equalize others. Along with this is the likely need to split systems out for their own comparisons, such as mechanical, electrical, telecommunications, and others so that they might provide a baseline on how much they each affect the overall PUE. Calculations for such determinations can become quite involved, but with the proper inputs and formulas we can be much better informed about how each affect each of the PUE results. Along with this other factors, such as cost, schedule, flexibility, reliability, and more can be overlayed for owners and operators to generate recommendations for short-term and overall data center benefit.
These calculations are not insurmountable; in fact, they are part of the services that experienced data center focused consultants should be able to generate for facilities, including all of the specifics that the PUE scores leave out, such as age, redundancy (tier), density, and more. The formulas behind them are still created from the basics of those given by standards and related groups. From all of this, the claims from vendors, designers, and others can be better understood instead of almost baseless claims of being able to lower PUE by a certain amount or percentage.
A more final resolution may be for the industry to centralize around the standard and guidelines that ASHRAE has already provided but adopt an approach similar to LEED in which every calculation should follow more detailed calculations that cannot be manipulated with the same defects provided by the current approaches. This will mean opening up on the design and efficiency approaches instead of being restrictive, then investigating the calculations for each project with respect to equipment, systems, and holistic design.
If PUE is going to remain useful for the industry, and not a method of manipulating numbers, the approach on how it is used and talked about will need to change. Just like greenwashing, PUE needs an image update along with how the calculations are performed. While it has been beneficial to push efficiency considerations and highlight overall energy use, the way it is misconstrued for competition and accolades should be reevaluated. Every time it is used, we should be asking whether it is another sales pitch or truly factual.
While PUE has played a significant role in advancing data center energy efficiency, it has always been a metric rife with limitations that skews accurate use. Its limitations, including the lack of consideration for renewable energy, variation with load changes, and neglect of other resource utilizations, highlight the need for a more holistic approach. By integrating additional metrics such as ERE, CUE, WUE, and IT equipment utilization metrics, data center operators can achieve a more comprehensive understanding of their facilities' efficiency and environmental impact. Embracing a multi-faceted approach to efficiency measurement will be crucial for data centers striving to meet the growing demands of a digital-first world while minimizing their environmental footprint.
