Updated: Sep 11, 2020
What is PUE?
PUE, which stands for “Power Usage Efficiency” is a data center oriented KPI that has been around since 2006. As it is about to enter its teen years, it is beginning to stir some perplexity as to its ultimate value as a KPI. But surely, it must have served some purpose as it has since become the most commonly used metric in the Data Center world to report energy usage efficiency. It was promoted by the Green Grid (a non-profit organization of IT professionals) in 2007, a group of highly knowledgeable professionals with a very noble mandate, to help us make better use of power, at least in data centers.
Why did it come about?
One thing that can be said about PUE is that it is an extremely simple Metric, made up of the ratio of just two values: Hydro Power over IT Equipment load. The first can easily be obtained from the utility company and the second can be read directly from the power plants used to supply the IT Equipment, be it a DC Plant or UPS. Not coincidently, This data was also easily accessible twelve years ago. Another fact about PUE is that as with most inventions, necessity was its mother. It quickly became clear that data centers were going to keep multiplying, that their power draw would also keep increasing, and, to complicate things further, so was the cost of Hydro. As with any “new” scenario, once the first designs have been laid out and the machine is put into action, always follows the “tweaking” period. “Okay, this sort of works… Now how do we make it better?”. Of course, the fresher the concept, the more things that can be done, fairly easily. A good analogy for this is the “Low Hanging Fruit”. Easy, high impact improvements that do not cost too much to implement. The PUE became the ultimate measurement tool in this phase of development of data centers. And it served this purpose very well.
Why question PUE now?
Besides the fact that an increase in PUE is bad, which can be counter-intuitive for some, this is a fair question since it served us so well for over a decade. Again the analogy of the low-hanging fruit comes into play. The early improvements were high impact, high visibility and therefore required little accuracy or precision to measure improvements due to an applied optimization initiative. Things have changed a low in the last 12 years, as they always do. The “easy to get to” fruit have been harvested. But since data centers are using more and more power, smaller percentile improvements can still mean considerable dollar value on the bottom line while also improving on the carbon footprint. The challenge that this reaching for narrower margins of improvements brings is that these are more difficult to measure. We are entering the realm of PUE “blurred lines”. Here is an example of this: A datacenter is retrofitted with new IT equipment that triples it processing power while only doubling its power consumption. Since the facility did not change in dimension, the cooling units may struggle more in dealing with the higher concentration of heat released by the equipment. This situation would yield a higher PUE, which is a bad thing. But in reality, the data center is using power more efficiently since it’s producing more intended work (processing) for less power as before. Another scenario is comparing the PUE of one site in Miami with that of a site in Alaska. Temperature and Humidity have a small but appreciable impact on cooling. So the exact same cooling equipment may be used in both sites, but one will have a slightly better PUE than the other. Similar effects can result from variances in utility power quality. So PUE is not necessarily a fair comparison from one site to the next depending on geographical location.
Is PUE on its way out?
With the advent of Big Data and Internet of Things, we can count on one thing; these things will both require a lot of new, powerful data centers, but they will in turn provide valuable data in order to more accurately measure their efficiency. We are entering the next phase in data center optimization. Reaching for the higher branches. New breaker panels, IT equipment power distribution modules are becoming smart. More and more, these are becoming accessible on-line. Temperature and humidity sensors are becoming standard items inside and outside data centers. Hydro power quality is being monitored. Whereas PUE incorporated every possible influence in power usage, we now have the data to break these into individual meaningful and accurate metrics. This means IT equipment power usage can be measured separately from the cooling systems efficiency. External influences such as geographical location can be taken into consideration and compensated for when measuring cooling equipment performance. This can be broken down much further, allowing to compare one piece of IT equipment to the next, which cooling equipment performs best for a specific geographical region, which geographical location provides high quality power at a lesser cost,… The new data analytics field of Machine Learning could also further improve on the past, detecting trends and relationships between various factors thus far missed by engineers. So what will happen to the PUE? It may end up carrying a different name, such as PURE “Power Usage Revised Efficiency”? Regardless, in the end, it will hopefully be made up of an aggregation of smaller but more accurate metrics into an ultimate KPI that will provide meaningful and fair representation of a site’s performance and efficiency at any given place and at any given time.
By Claude Morin, Facilities Management and Subject Matter Expert Missing Link Technologies Ltd.