One of the most frustrating things about building energy efficiency into data center infrastructure is that it is very difficult to determine how successful you’ve been.https://o1.qnsr.com/log/p.gif?;n=203;c=204663295;s=11915;x=7936;f=201904081034270;u=j;z=TIMESTAMP;a=20410779;e=i
Measurement techniques and benchmarks abound, to be sure, but the problem lies mostly in the interpretation of the data, rather than in its collection and categorization. Take the Power Usage Effectiveness (PUE) metric as an example. The granddaddy of the green data center movement, PUE is designed to calculate energy efficiency by measuring the amount of power going to IT equipment vs. that going to facilities-related infrastructure like cooling, lighting, etc. Many organizations tout their recent reduction from ratings in the 2.0 range or higher (meaning only half or less of their consumption went to actual IT equipment) down to 1.5 or less, which in some cases would indicate a near-doubling of energy efficiency over the past decade.
When you parse the numbers, though, a different picture emerges—one in which the potential exists for organizations to claim broad success based on operational ideals rather than real-world usage (welcome to the world of benchmarks.)
As researchers at Stanford University found out recently, one of the chief flaws in the way PUE is implemented is that it provides a one-time, instantaneous measurement of energy consumption. This makes it easy for operators to report the minimum observed values within a given testing period. A far more accurate practice would be to calculate the average PUE over a length of time—the researchers suggest a year—and then compare the results to a set of established sustainability and energy consumption codes.
Another problem is the fact that many data centers share facility space with other entities, namely office space, labs or even retail/showrooms. According to IO Datacenter’s Patrick Flynn, PUE makes no distinction between stand-alone or mixed-use facilities, nor among relative sizes, designs or even the data loads they are required to process. That’s why IO is designing its newest facilities around modular data infrastructure featuring real-time PUE measurement. Not only does this provide a working average of PUE over time, as recommended by Stanford, but it provides a much more granular view of the IT/energy paradigm by allowing users to view their energy consumption patterns right down to individual data modules.
The shortcomings of PUE have been an open secret in IT circles for some time, which is why the standard has generated a number of sub-metrics designed to help parse the data. These include “container-only” or “cooling-only” versions, according to Denis Murphy, CEO of Irish automation systems developer Anaeko. Of course, this does little to prevent organizations, especially cloud providers, from presenting their facilities in the best light, particularly in the absence of globally recognized certification processes that would apply to centers in Ireland versus those in, say, India.
Still another problem is that, as a nearly 10-year-old standard, PUE is starting to lose its cachet in modern, consolidated data centers, according to Sev Onyschkevysh, CMO of data center management firm FieldView Solutions. As densities, and therefore utilization, in the server room and elsewhere increase, active systems will run hotter, which causes legacy cooling infrastructure to work overtime in order to keep the thermostat down. Ideally, the solution is to upgrade cooling infrastructure to more accurately address the needs of consolidated environment, but in the meantime PUE will likely increase even though the data center is running on a reduced hardware footprint.
All of this points to the complexity associated with environmental issues. It seems that advancements in one area’s “greenness” have a detrimental effect somewhere else. Updating measurement techniques and benchmarks is a worthy goal, as is applying them as universally as possible, but there will probably always be a way to tweak the numbers in someone’s favor.
For data center operators, however, one metric is incontrovertible: If your energy bill is going down, you’re probably doing something right.