PUE is about Efficiency, Not Performance or Cost
April 6th, 2012 By: Industry Perspectives
Derick van Heerden is a senior IT architect in IBM’s Integrated Technology Services division, located in IBM Cape Town, and provides consultation on data centre efficiency, capacity and resiliency as well as facility management.DERICK VAN HEERDEN
Management of a data centre’s Power Usage Effectiveness, known as PUE, is not a strategy for improving the performance of the data centre or reducing data centre power consumption. PUE (or more accurately, its reciprocal, DCiE) is a measurement of efficiency. It does not measure data centre performance or power consumption and should not be expected to do so.
Efficiency is a measurement of the economy with which a goal is accomplished. It measures the ratio of output to input. It follows that the numerical value of an efficiency measurement falls in the range of 0 to 1 and is commonly expressed as a percentage. So, if the output equals the input, the efficiency of the process is 100 percent.
In the context of electrical engineering, the currency of efficiency is power, measured in Watts. Why? Power is the product of two other elements of electricity that can vary widely from input to output: voltage and current, and thus the only element that can be measured on the input and output sides of the device in a meaningful manner. Data Center infrastructure Efficiency (DCiE), and its more famous reciprocal, PUE, applies this understanding of efficiency to data centres. Thus, PUE measures power efficiency in the data centre. When the measurement moves from instantaneous to time-bound power usage, PUE becomes a proper measurement of data centre energy efficiency.
PUE and Data Centre Performance
Efficiency expresses the economy with which a device transforms input to output power. It is completely agnostic about the magnitude of the device’s input and output power, as well as the device’s function. So, whether the device is pumping water or flipping bits is irrelevant to determining its efficiency. The moment the concern shifts to the relationship of a device’s input power to its function, the device can no longer meaningfully be assessed in its own right. It now has to be compared to other, similar devices to determine whether it is any good. The focus shifts to determining which device is best at performing a particular function, and in doing so, it shifts from talking about efficiency to talking about effectiveness, which is a measurement of the success with which a goal is accomplished.
By linking the functional outcome of a device to its input power (measured in Watts), we have potentially found a metric for measuring effectiveness (or performance). This may well be something worth having in the case of electrical devices that perform a single function. However, its usefulness diminishes when a device, such as an IT server, can perform a multitude of functions. An analysis of all the different potential functions a server can perform reveals that, in the end, it all comes down to the manipulation of bits. But the manipulation of bits is as abstract an activity as it is useless at expressing a meaningful notion of performance or productivity.
PUE and Energy Cost
A decreasing PUE signals an increase in the efficiency of a data centre. It does not suggest that less power is used; merely, that whatever power is used, is used more efficiently. The expectation that PUE should fall when IT power consumption is reduced as a result of virtualization and consolidation efforts is therefore wrong. Reducing only the IT load means the ratio of input to output power has changed for the worse, and contrary to some people’s expectation, the data centre has become more inefficient: a proportionately larger part of the input power is expended on the non-IT workload, notably cooling.
A far more significant effect is achieved if the power consumed by the non-IT load is reduced. Such a reduction means that a greater proportion of data centre input power is available on the output side (the IT load), leading to greater energy efficiency.
Thus, to summarise. PUE is a metric of efficiency. It does not measure performance or power consumption. But that does not mean that PUE is deficient. Conversely, it does not mean that performance and power consumption are aspects of energy efficiency. These are separate aspects of data centre operations and must be measured by different metrics. The easy one is power consumption, since it is a measured by kilowatt-hours consumed. More difficult is data centre performance for which no satisfactory metric has yet been proposed.
Industry Perspectives is a content channel at Data Center Knowledge highlighting thought leadership in the data center arena. See our guidelines and submission process for information on participating. View previously published Industry Perspectives in our Knowledge Library.
FANTASTIC piece, Derick. Your attention to detail and the mathematics of PUE and DCiE make a convincing and persuasive argument. Reading this got me thinking, why is it that PUE is used so often for something it really doesn’t do, then? Rather than leave an enormous comment here I posted a blog piece on IT toolbox (http://it.toolbox.com/blogs/hardware-maintenance/pue-is-about-efficiencyso-why-is-it-used-to-measure-costs-and-performance-51008), the gist of it being that ‘efficiency’ can be expressed mathematically while ‘cost’ and ‘performance’ are more subjective. The appealing PUE gets shoehorned into giving cost and performance the semblance of objectivity when vendors are selling product and when IT managers are supporting a proposal for change.