The Problem with PUE Metrics

The Green Grid releases a new metric called PUE (power use effectiveness). Is this the end-all be-all metric for data centers?

In 2007, the Green Grid released a metric by the name of PUE, otherwise known as power use effectiveness, in which data centers use to measure energy efficiency. The problem is some fudge the numbers in order to appear more efficient than they really are. Is it fair that PUE be used as the end-all be-all metric for data centers?

PUE – The Dominant Metric

In recent years PUE has surpassed all other measurement standards in term of popularity. Competing data center companies have a tendency to one-up each other so-to-speak.

In short, PUE represents the ratio between power into the facility Vs. power consumed by server equipment. Basically, data centers want to have a close ratio between the two, indicating that very little power is lost in the delivery chain.

Many seem to use PUE as good PR, which isn’t a bad idea if you’re trying to appeal to the energy conscious. It’s really easy for companies to report numbers that aren’t really close to what their true PUE is. For example, a data center will run much more efficiently in cold weather than in hot. Why? Because it’s easier to pump in cold air from outdoors, allowing air-conditioners to run less. That, right there would deliver a lower PUE. But therein lies one problem. As energy efficiency increases, so does a facility’s PUE rating. And a higher rating would look bad to customers, who don’t realize it’s all about the ratio (power in vs power consumed). So what’s the industry to do?

Bring Back other Metrics

There are other competing metrics but the fact that everyone is using PUE as the benchmark means that its tenure is likely to stay. But what if one other metric could be used to supplement or offset standard PUE ratings? David J. Cappuccio, Research VP at Gartner proposed an interesting metric last year. He calls it RUE or rack use effectiveness. After-all, the server rack is where most important hardware resides and since all the power systems in the data center are meant to support said hardware wouldn’t it make sense to measure efficiency of the rack itself? You can read more about his proposal by clicking here. Brace yourself though, it’s choked full of some mathematical equations that might just distract you from your morning routine. Nevertheless, Cappuccio makes a very fine point: The data center needs more “usable” metrics in order to offset this dependency on PUE and get closer to the truth.