The Power Usage Effectiveness (PUE) metric is the de facto measurement when it comes to the efficiency of data centres, but its days could be numbered.
The news earlier this month that the UK Environment Agency will be one of the first customers of Capgemini’s new Merlin data centre, which boasts a PUE of just 1.08, is the most recent evidence of the metric’s preeminence – but moves are still afoot to supersede it.
Essentially a ratio, total data centre power divided by IT equipment power, PUE is actually a pretty crude tool but a relatively popular one. The closer to 1 the result, the more efficient the data centre is as more of the power entering the facility is getting to the servers.
Formulated by the Green Grid, a consortium of vendors including IBM, Dell and APC, and some end-users, the metric has served as a satisfactory stop-gap but some experts, including the Green Grid itself, are questioning its future.
first conceived in 2007, is PUE already over the hill in 2010? It could be a victim of the success of the green data centre movement.
Rising energy costs, an expanding data centre market and increasing efforts by governments to control energy and carbon use, are all fueling the need for more accurate measures. Legislation such as the UK Carbon Reduction Commitment (CRC) is already creating a stir amongst the countries DC operators. When it comes into force next month, it will require data centre owners and operators to report and curb the energy used and carbon produced by their facilities.
Ultimately a product of a vendor-backed group, the idea of integrating PUE into legislation, along with its reciprocal known as Data Centre Efficiency (DCE) metric, has some industry insiders concerned. The Green Grid itself concedes that the metric has no comparative value and should only be used as an internal indicator of overall data centre efficiency. But PUE continues to gain traction and in June this year was adopted by the US EPA’s Energy Star rating body as its metric of choice.
The Green Grid is taking the evolution of an IT productivity metric seriously and has a team of around 20 dedicated to the task. Their most recent output has been focused on effectively slicing PUE into a subtler and more accurate measure. In July 2010, the team proposed four new sub-categories of PUE which range in detail according to the sophistication and capabilities of the DC operator in question.
But while modifying PUE has its benefits, the holy grail of data centre metrics is one that crucially takes account of IT productivity or “useful work’ done by components such as servers – something PUE effectively ignores. As the originator of PUE, The Green Grid has taken it on itself to suggest its productivity-based successor. This is obviously more than altruistic, given that the resulting metric could be integrated into environmental legislation which will impact its members.
The proposals put forward by the Green Grid are characterised under the umbrella term of DCP or the Data Centre Productivity metrics. This family of measures is essentially defined by the “useful work” produced in the data centre divided by the resource consumed to produce that work. The resource in question is open to interpretation but the Green Grid is pushing ahead with energy as a likely contender with DCeP the resulting metric.
But while DCP may make sense on paper, applying it to a real-world DC environment has proved problematic. There is no industry, and little academic, consensus on how to effectively measure the “productivity” of an IT system. Coming up with a common measure for the myriad categories of “useful IT work” has been written off by some experts as a fool’s errand and a distraction. Proposing to then tie such an elusive measure to data efficiency could introduce levels of complexity which would make any model practical useless as a comparison tool and a woeful instrument for environmental legislation.
Alternatives to PUE do exist such as Emerson Network Power’s CUPS metric, CADE – developed by consultants McKinsey in partnership with The Uptime Institute, and DPPE from Japan’s Green IT Promotion Council. However none of these has the consensus that PUE enjoys which is perhaps the most crucial issue. Accuracy is ultimately trumped by ease of use for many DC owners.
However, groups such as the British Computer Society’s Data Centre Specialist Group (DCSG) believe the future lies not in chasing the Snark of productivity but taking a more holistic approach to IT efficiency. The group drafted a paper in August this year to that effect and has submitted it to the Green Grid and other bodies for approval. The somewhat myopic approach typified by PUE, and currently favoured by regulators, should be seen as stepping stone towards metrics, and an associated regulatory regime, that looks at IT efficiency in the context of the whole business.
This holistic approach to the intricacies of data centre efficiency is certainly valid and an admirable goal, but the increasing legislative burden on DC operators means that workable metrics are needed now. While it is true that a productivity-based IT metric is wanted and required by data centre operators keen to optimise their facilities, pressure from outside may prove more decisive. Governments are keen to regulate, and ultimately tax, the considerable carbon contribution of the data centre sector and need industry-agreed metrics to do so.
While factoring in productivity metrics may ultimately result in fairer laws, consensus may well trump accuracy in this scenario. Ultimately, evolving PUE into a more sophisticated tool, in line with the current efforts of the Green Grid, makes sense until something better comes along.
PUE is far from perfect and risks being subverted into a meaningless marketing tool. But if it can be protected and nurtured it may continue to prove useful to regulators and industry alike. Finding a really good IT data centre metric is a quest worth pursuing, but with spiraling energy prices, and ever restless regulators, data centre operators will have to endure the consolation prize of PUE for now.
US widening lead over China on AI development, as UK places third in Stanford index…
Amazon to invest a further $4bn into AI start-up Anthropic, doubling its investment as it…
The demand for tech skills is surging, driving economic growth but revealing challenges. Financial costs,…
US Supreme Court tosses Meta's appeal over Cambridge Analytica-linked investor lawsuit, meaning case must proceed
Uber reportedly seeks $10m stake in Chinese autonomous driving firm Pony AI via US IPO,…
iPhone maker reportedly developing next-generation AI large language model for Siri for spring 2026 as…