Managing Your PUE

April 3, 2012 3 Comments »
Managing Your PUE

PUE (power usage effectiveness) is the quintessential metric for data center efficiency. Seeking a good rating by this metric can earn your company not only marketing points, but savings in energy bills as well. But shooting for a low PUE value isn’t the end of the energy efficiency story.

PUE and What It Means

Power usage effectiveness (PUE) is simply the ratio of total data center energy consumption to the amount of energy consumed by the IT equipment alone. Since the data center’s primary purpose is to supply IT resources to a company or consumers, an ideal PUE is 1.0: all energy consumed by the data center goes to the IT equipment. That would mean no energy is used or lost in the power distribution system (including cables, uninterruptible power supplies, voltage conversion stages and so on), security system, lighting, heating/cooling or any other non-IT portion of the data center. Obviously, then, a PUE of 1.0 is unattainable, since it requires 100% efficiency.

Of course, simply looking at the definition of PUE leaves some grey areas. What if a company has solar panels (or some other on-site power generation infrastructure) at its data center campus, for instance? Does that power count toward the total power consumed by the facility? Should peripheral power drains like lighting and security really be included in the PUE calculation? The Green Grid, the organization that developed the PUE metric, provides some guidance on how it is to be properly calculated, but unscrupulous companies can easily “cook the books” to make their facilities look more efficient than they really are.

Thus, PUE measurements should always be taken with a grain of salt, particularly when no details are provided regarding how they are calculated. If you ever see a PUE of exactly 1.0 or less, you can be certain you’re being lied to. But what about a PUE of 1.001? Believe it only if you think that only 0.1% of every watt delivered to the data center’s servers is lost in power distribution inefficiencies—to say nothing of lighting or anything else happening in the facility. In other words, don’t believe it. But for a company honestly seeking to improve its energy efficiency (not just its public image), PUE can be a helpful metric. Here’s some ways to manage it.

What You Don’t Measure Won’t Help You

To know your PUE, you must measure it. Unfortunately, looking at the power rating labels on equipment just isn’t enough—most equipment, including IT systems and cooling infrastructure, varies in its power consumption throughout the day and over the course of the year. The following are some considerations to aid in PUE measurement.

  • Measure IT power as close to the source as possible. Unless you’re using zero-resistance cables (believe me, you’re not), then the farther you go from a server, the higher the power consumption reading for that server—and the lower your PUE. That may be good for marketing, but it’s not good for an honest PUE assessment. When you take measurements close to the IT equipment, you’re getting a more accurate reading of the IT power consumption. That may hurt your PUE calculation a little, but it gives you a more honest assessment of your data center’s efficiency.
  • Don’t cut corners on total power consumption. Assuming you don’t do any on-site power generation, the best way to calculate your total power consumption is to just look at your energy bill. See what the power company says you’re consuming, and use that number. (Yes, I suppose the power company could be wrong, and it may not hurt to check once in a while, but this approach reduces the temptation to fudge the numbers.) Don’t cut out security, lighting and other needful systems—but if you must, make note of it if you report your PUE publicly. Of course, you’ll need to do some actual measurement of your own to calculate PUE on a finer-grain time scale, but the point is this: don’t pick and choose systems and then say “these don’t really count.”
  • Measure often and regularly. The closer you can get to a continuous PUE measurement, the more information you will obtain to help you improve efficiency. PUE varies daily depending on workload, environmental conditions (outside temperature) and other factors. Furthermore, it also varies over the year, largely for similar reasons. To get the best sense of your data center’s efficiency, measure PUE as often as possible in a regular manner—don’t skip certain times of day or year (another way to fudge the numbers).

Improve Your PUE

Once you are able to measure your PUE, you can work on improving it. (Even if you don’t care about PUE, you can still improve your data center’s efficiency; but if you want numerical data to back you up, here’s some steps to take.)

  • Hot/cold air isolation. One of the best things you can do in terms of maintaining good airflow in the data center is to isolate hot air from cold air. A larger temperature differential improves cooling efficiency. How you implement this isolation depends in large part on your cooling approach and infrastructure and your equipment layout.
  • Raise the temperature. Many data centers operate their facilities at temperatures much lower than the maximum recommended by ASHRAE. By raising the thermostat, you automatically save cooling costs.
  • Use free cooling when possible. Why use mechanical cooling when outside air is often cool enough? Free cooling (although not entirely free) can slash power consumption by cooling infrastructure tremendously—and thereby significantly lower PUE.
  • Focus on power distribution. Using high-efficiency UPS systems and eliminating unnecessary voltage conversion stages reduces power loss. (And since this power loss becomes heat, it also reduces the cooling load, meaning even more savings.) After cooling, power distribution is the primary leech on your efficiency.
  • Implement controlled lighting systems. Employees may not be able to function in the dark, but servers can. So don’t waste money and power on lighting when no one’s in the facility. Consider installing a lighting control system that shuts the lights off automatically at certain times or under certain conditions (such as when no one is in the facility).

Don’t Stop at PUE

PUE provides some indication of the efficiency of your data center, but it is not the end-all be-all of efficiency metrics. Imagine you went into your facility today and replaced all the servers with less efficient models: your PUE would improve! By the same token, installing more-efficient servers hurts your PUE. Don’t ignore IT efficiency just for the sake of PUE (you are billed by your utility according to how many watts you consume, not according to your PUE). You may even want to measure and seek to improve your data center according to other metrics as well—and numerous ones are out there. Yes, you want to minimize the amount of power your facility consumes in peripheral tasks (power distribution, cooling, lighting and so on), but you also want to minimize the amount of power your IT equipment consumes. In other words, the ideal facility should accomplish the maximum amount of work using the minimum amount of power. And numerous benchmarks are available to measure performance per watt—although these metrics also can be manipulated and must be used with care, just like PUE. So, PUE is helpful, but it isn’t the end of the story.

Conclusions

You can take steps to obtain useful and accurate measurements of your data center’s PUE, and once you do, you can seek to minimize it through a variety of steps. But don’t stop there—PUE is useful, but it doesn’t deal with all aspects of data center efficiency. IT equipment—the biggest energy consumer for data centers with a PUE below 2.0—can also always stand to see greater efficiency. Supplementing PUE with other metrics can help avoid a myopic view of energy efficiency, enabling companies to garner the maximum savings from their efficiency improvements.

Author contact

Photo courtesy of mastermaq

About Jeff Clark

Jeff Clark is editor for the Data Center Journal. He holds a bachelor’s degree in physics from the University of Richmond, as well as master’s and doctorate degrees in electrical engineering from Virginia Tech. An author and aspiring renaissance man, his interests range from quantum mechanics and processor technology to drawing and philosophy.

Pin It

3 Comments

Add Comment Register



Leave a Reply