At the 2011 Uptime Institute Symposium in Silicon Valley in May — one of the major international conferences on the data center calendar — the audience witnessed an impassioned debate about one of the critical elements of modern IT: data center energy efficiency.

Experts from organizations dedicated to reducing IT energy wastage — the Green Grid, the Silicon Valley Leadership Group, the Climate Savers Computing Initiative, the Critical Facilities Roundtable and the Uptime Institute itself — went head-to-head on how CIOs can best demonstrate the efficiency of their IT powerhouses.

From some, such as Pitt Turner, the executive director of the Uptime Institute, there was surprising advice. Rather than 
use existing metrics such as Power Usage Effectiveness (PUE) or wait for the industry to come up with new and better measures, data center owners should find ways of doing it themselves.

Make up your own metrics? What exactly is going on — or indeed wrong — in the world of data center efficiency measurement that an organization such as the Uptime Institute is giving such advice? And what could have shaken faith in the PUE metric that is already so widely used?

The answer to that, like almost everything to do with data centers, is a mix of the simple and the complex.

Basic measurements

The simple story is that the industry has had a good metric since 2007, when the US Environmental Protection Agency produced 
a seminal report on data center power usage, concluding that a large portion of the vast — and increasing — quantities of electricity used by data centers is wasted. In that year, the Green Grid was formed and produced PUE — a metric that is easily calculated and understood at all management levels.

PUE compares the total power consumed by a data center to the power consumed by one component part, the IT load, so that any overhead or waste can be identified (the equation is: total load/ IT load). So a data center drawing 1.6MW of power of which 
1MW is actually consumed by IT workload has a PUE of 1.6, for example. As testament to what has been achieved in recent years, data centers in 2007 typically had PUEs of 2.3 or above, while recent research suggests PUEs have fallen to somewhere around 1.8.

The PUE metric is now the defining metric for data center energy efficiency. It has given management a simple vocabulary for talking about energy and data centers, and has helped operators identify and reduce inefficiencies. Its simplicity also means there is a single number to aim at, to compare, and, if possible, to publicize.

Misapplying the metric

This is where things get complicated. When PUE was introduced, it was understood to have a limited role: to measure the electrical efficiency of the data center infrastructure that houses the IT equipment in a single facility, over time. It was never intended to compare two data centers, which could never be similar enough to support the equation.

But out in the wild, it has become a different story. PUE figures are cited like golf handicaps. Investment analysts looking for indicators 
of inefficiency are citing them in some procurement documents, and they are appearing in planning rules for new data centers and in the bills of service companies.

All of this rightly concerns those facility engineers who originally devised PUE and similar metrics. Not only is the metric being widely misapplied, but there is little recognition of just how malleable and site-specific the numbers are. PUE is highly sensitive to a wide range of variables such as the amount and type of IT equipment in the facility; the age of that equipment and the facility; the type of cooling employed; the location (cooler weather usually means less cooling); humidity; and, importantly, the data center’s resiliency (greater resilience can mean a greater energy overhead).

And there are further issues around when and how energy measurements are taken. Rather like the price/earnings value of a listed company, PUE comes in many flavors: prospective PUE; design PUE; PUE levels 1, 2 or 3; average PUE; trailing 12 months PUE; forward looking PUE; capacity PUE; peak PUE; or even, as is currently gaining ground, “true PUE,” in which onboard systems cooling (i.e. fans) is counted as infrastructure overhead, making the data center look much less efficient. And most recently, with PUE 2, the Data Center Efficiency Task Force has tried to tighten the measurement requirements of the metric.

But PUE is not even the main issue; its limitations are well understood. A bigger issue is that one data center, using older servers inefficiently, may have a far better PUE figure than another that is using modern Energy Star-rated servers, a lot of virtualization, and much less energy overall.

Greener estates

This seems to call for a metric that takes into account the useful work done by the data center, and the total energy used to do this. But what is useful work?

A supercomputing center might be able to measure its work in floating point operations per watt, and an ecommerce company in Java transactions per watt, but what about a company with hundreds or even thousands of different applications with different requirements? It has been proposed many times before that perhaps companies, or even industry groups, could come up with their own metrics — banking transactions per watt, or search queries, messages sent, or minutes of phone calls connected.

This is the area that the Uptime Institute’s Turner was alluding to — each business knows what it wants and needs to measure, and probably has a lot of the data to hand. The Green Grid’s experts, in fact, have done half the work: they came up with a metric (the “productivity metric”) as far back as 2008 into which any definition of work can be inserted to produce an overall measure of productivity. But it has proven too complex, as did an alternative — known as CADE — put forward by Uptime and management consultancy McKinsey. Since then, the search has been on for something simpler.

Some CIOs are beginning to look at this in another way: if they can plot the energy efficiency of all their IT equipment against a best possible ideal (such as the most efficient servers on the market), and they can further measure the utilization of that equipment along with efficiency of the data center itself (PUE), they may be able to come up with one percentage number — an efficiency number that maps itself against an ideal. This number, fairly easily collected by the IT department, would rightly be open to all kinds of criticism but it might just be the approach to trigger a new wave of competition as CIOs seek to run a greener estate than their rivals.

Further reading: The changing shape of the corporate data center

As research director for data center technologies and eco-efficient IT at analyst company The 451 Group, Andy Lawrence advises businesses, public sector organizations and IT suppliers on their green and data center strategies.