Ever since the PUE metric was originally developed by The Green Grid in 2007 it’s been widely adopted in the Data Center industry. Yet it’s flawed in so many ways, in this blog we examine some of the flaws and challenges to PUE, and how Gaia Edge is taking a new, holistic approach to PUE in light of key sustainability.
Where PUE (Power Usage Effectiveness) is concerned we’ve long poured over ways, means, technologies, and methodologies to get to the magic ratio of 1:1, so much so in fact that Google now boasts an impressive 3.3 Million search results for Data Center PUE alone. Everyone from Google, Gartner, The Uptime Institute and Statista, not to mention vendors and tech providers have been weighing in on the subject of PUE and of course ways of decreasing or improving current PUE’s. Here’s just some of the cloud of what they have to say;
Statista on Power Usage Effectiveness - In 2020, the average annual power usage effectiveness (PUE) ratio within the largest data center of each of the respondents was 1.59.
Google on PUE’s within their Data Centers - Our best site could boast a PUE of less than 1.06 if we used an interpretation commonly used in the industry. However, we're sticking to a higher standard because we believe it's better to measure and optimize everything on our site, not just part of it.
Gartner on PUE- “There are five key steps data center managers should follow in sequence to maximize efficiency, as well as a ‘Step 0’ – measuring and monitoring power usage – so that improvements can be monitored along the way.”
With so much noise information around PUE across the industry and big strides being made (if we take Statista data) from an average PUE of 2.5 in 2007 to a current average of 1.59 we can’t help but wonder if PUE is still relevant and if not what are the alternatives?
PUE Alternatives to Data Center Efficiency
The short answer is of course PUE has a place in today’s Data Center, as a widely implemented, industry-standard measure PUE has clearly had far-reaching impacts on the measurability and accountability of the effectiveness of power usage across environments. However one of the downsides to PUE is, by the very nature of leveling the benchmark across all data center environments, is that it obscuring the actual energy use of some of the industry’s biggest consumers.
We muse that the public and enterprises may feel comfortable and comforted by knowing that their estate is running at better than average PUE’s maybe on or around 1.2 for net new estates but what about if we flipped that into real world figures, maybe 40 mW, 40,000 kW or 40 Million Watts, enough to make 1.28 Million cups of tea? Perhaps putting it plainly, how would we feel knowing that our data centers use enough energy to power a small country?
Moreover there a basic, obviously flaws not to the PUE calculation but to the limits and scope of what is being calculated. In our latest research “The Race to Zero” we spotlight the following example:
Overall Energy use 9872 KwH
Overall Energy use 7532 KwH
Clearly DC2 has either more efficient computation infrastructure or less efficient cooling and facilities infrastructure, it could be argued then that whilst DC1 will claim better energy efficiency and boast better PUE’s than their counterpart their overall consumption is very clearly more harmful and emitting more carbon – spotlighting a fundamental flaw in using PUE in isolation.
Data center leaders are also subject to the same issues as they seek to make the very improvements that PUE is designed to affect. For example, a data center server stack using for example 300 kWh of power daily is submerged using the latest in immersion cooling technology, this reduces the amount of energy needed to power the hardware. Whilst this should be a positive step, what this actually means for PUE is that the reduction in compute power generates a higher PUE when compared to the same non compute power, a paradox then for anyone looking to make significant energy improvements whilst adhering to a benchmark industry standard.
Whilst there is no denying that PUE has been important for improving the efficiency of the ways that the data center industry uses power there are growing calls for this measure to be updated and integrated with more sustainability based measures. We examine below just some of the other metrics available.
PUE Alternatives to Data Center Efficiency
Performance Per Watt is hailed in some corners as a potential alternative to PUE, for those unfamiliar with the concept Performance per Watt or PPW is calculated as follows:
(PI * Avg Device Utilization / Watts ) * 100
Where PI is the relative performance of the device in question. Whilst we see technology vendors actively using this metrics there are no known examples of full facility data centers using Performance Per Watt as an active measure. However, it does raise questions around whether PPW could be implemented across whole estates to identify optimal and lower performance hardware that could result in whole scale improvements.
In the context of PPW we mustn’t fail to mention Koomey’s Law and the trends across processing and compute environments. Koomey’s Law describes a trend in the number of computations per joule of dissipated energy. This number doubled every 18 months from 1945 to 2000 (100x per decade), then slowed, doubling every 2.6 years or so since (16x per decade). In other words, what we’re currently seeing is a trend for the amount of power needed for compute processing halving every 2.6 year, so surely we should be seeing much better PUE than 1.59?
If we lived in a static world then yes, of course we would have seen the PUE barrier shoot down below 1 but Koomey’s Law is subject to the same voracious appetites for data, connectivity and performance that we’re all subject to and whilst then technologies are accelerating so too is our consumption – with the two vying for first place.
As with Moore’s law we all know that there is only one long term outcome for trends like Koomey’s law and that is a plateau – what that means for data centers and of course, anyone passionate about sustainability, is that we have to find another way.
PUE Alternatives to Data Center Efficiency
A relative newcomer to the Data Center world, DCEP = Useful Work Produced / Total Data Center Energy Consumed over time. Also pioneered by the Green Grid, DCeP effectively looks at the relative value of useful work, for this example, we could consider “useful work produced” active compute functions in specific time frames versus the overall energy usage of the data center.
In reality, the proposal looks a little more tricky but the Green Grid proposes the following equation for useful work;
(Sum of all tasks * Value of the task) * Time-Based Utility Function * Absolute Time of Completion
In simplifying this task values can be broadly assigned as 1 – giving no weighting to the value of the task itself, whilst this does simplify the calculation it also results in potentially huge inconsistencies across data center environments as each facility and/or business assign a different weighting to “useful work”. It was hoped back in 2014 when the metric was established that over time measures would harmonise to form a more comparable benchmark but 7 years on it seems that DCeP is still largely unused.
A new data center metric on the horizon
In conclusion, it seems that none of the metrics, proposed or in use provide a genuinely useful measure for our sustainability goals, but by combining metrics with fundamental environmental, carbon, and sustainability measures organisations can truly impact their own data center efficiency metrics.
At Gaia Edge, we believe that Zero Carbon data centers are the future and that we must act now to drive measurable performance improvements and commercial goals with sweeping reductions in our Carbon footprint. As we strive for better PUE, DCIE, DCeP and other measures data center leaders must acknowledge that the efficiency of energy use is only one part of the story.
We must go further than “how effective is our power use” to look at the fundamental use itself, actively seeking alternatives to traditional fossil fuels, proactively reducing the amount of energy we do use, and of course, steadfastly moving towards green, sustainable energy sources. Moreover, as we move into what the UN have termed the “decade of action” more than ever we have a responsibility as an industry to be clear, open and transparent with our efficiency and consumption measures.
19-20 Bourne Ct, Redbridge, Woodford Green, IG8 8HD, United Kingdom
Copyright © 2021 Gaia Edge Ltd - All Rights Reserved