Data Center Journal

Volume 30 | February 2014

Issue link:

Contents of this Issue


Page 25 of 32

THE DATA CENTER JOURNAL | 23 t he metric that probably comes to mind first is power usage effectiveness (PUE)—a de facto industry standard for measur- ing the energy efficiency of a data center. But given its limitations, PUE isn't enough; it doesn't always accurately convey efficiency information, and it doesn't measure other relevant operating parameters like water consumption, per- formance delivered and so on. But just be- cause your car has a speedometer doesn't mean there's no room for a tachometer. So what might a data center "dashboard" include to give data center managers an accurate picture of what's happening? poweR usage effeCtiveness Energy is the greatest ongoing cost of operating a data center. e cost of power over a server's lifetime, for instance, typi- cally exceeds the capital cost of the server, leading to a total cost of ownership greater than twice the purchase price. Controlling energy consumption is helped tremen- dously by measuring and tracking it. PUE is one metric that can help: it is simply the ratio of power used by the entire facility to power used by the IT equipment. A value of 1.0 is ideal, indicating that power is only consumed by servers, networking and so on rather than by cooling, lighting and power-distribution inefficiencies. A recent Digital Realty Trust survey indicated that the average data center PUE for large com- panies in North America is 2.9, meaning almost two out of every three watts con- sumed by the data center go to something other than the IT equipment. A companion metric of PUE is data center infrastructure efficiency (DCIE), which is simply the inverse of PUE multi- plied by 100%. So, for instance, a PUE of 2.9 is equivalent to a DCIE of about 34.5%. DCIE values may be slightly more intui- tive, with 100% being a "perfect" score. Because it is essentially just a mathematical alteration of PUE, it suffers from the same problems of PUE. Ignoring attempts by some compa- nies to game the system, one of PUE's main drawbacks is its lack of any performance input. For instance, if a company replaces its servers with new, energy-efficient mod- els, its PUE can actually rise even though the facility is operating more efficiently. In addition, because of the limited factors that go into calculating PUE, it is a poor choice for comparing different data centers. Nevertheless, in a given data center, PUE (or DCIM) can be a valuable overall metric for self-evaluation and tracking progress as well as variation in efficiency at different times of the day or year. MeasuRing poweR To calculate your facility's PUE, you need to first gather power data. Any num- ber of data center monitoring products en- able managers to collect this information, but a number of considerations arise as to what exactly goes into PUE. For instance, where is power measured? Measuring pow- er consumption at each server may or may not be practical, but the further away the measurements from the IT equipment, the less accurate the final calculation will likely be. Furthermore, frequency of measure- ment is critical: a facility's power consump- tion will vary depending on a number of factors ranging from time of year (varying weather) to time of day (changing work- loads) and so on. In addition, you need to decide what in particular to measure, particularly with regard to peripheral systems such as lighting. Arguably the most "honest" way to measure total data center power is to simply look at kilowatts (or megawatts) at the facility's service entrance, but if your data center is part of a larger building housing other departments, this simplistic approach may not work. Since PUE should just be a "self-check" metric rather than something that is advertised or used to compete with other data centers, the minor details of what counts toward data center power and what doesn't are of secondary importance; the key is consistency, what- ever your choice. otheR effiCienCy MetRiCs PUE is by no means the only power- efficiency metric for data centers, but it is the most well known. Additional ones have been proposed and are in use. Beyond power, however, are efficiency metrics for other data center resources. e Green Grid (which developed PUE) has also developed metrics for water and carbon. Water is slightly more tangible, as it is a resource whose usage can be measured by the data center operator. In particular, water usage effectiveness (WUE) is the ratio of water used (in liters) to energy consumed (in kilowatt hours). Obviously, the less water used the better, so the ideal in this case is a WUE of 0. Carbon usage effectiveness (CUE) is more abstract, as a given company has Imagine driving a car with no indicators or gauges on your dashboard. Now, imagine driving a car with an abundance of useful gauges. Information like speed, engine RPMs, coolant temperature, how quickly you're using fuel and so on all help ease the task of getting from here to there by letting you know what's happening under the hood. The situation in the data center is much the same: controls that measure, process and deliver operating information give data center managers the ability to better govern operations for optimum efficiency, performance and uptime.

Articles in this issue

Links on this page

Archives of this issue

view archives of Data Center Journal - Volume 30 | February 2014