Microsoft Eliminates Server Fans, Despite PUE Hit

Not all steps that improve the energy efficiency of your data center will boost performance in key metrics for measuring “green” data centers. An example: removing fans from servers, which actually has an adverse effect on Power Usage Effectiveness (PUE), the leading metric for data center efficiency.

This dilemma was discussed in a new video from Microsoft featuring Dr. Dileep Bhandarkar, Chief Architect for Microsoft Global Foundation Services, which designs and builds Microsoft’s data centers. Bharkander discusses how Microsoft realized it was duplicating effort by using fans in both servers and air handlers in its data center environment. So it designed the latest version of its IT-PAC data center container module as a giant air handler, with airflow management that eliminated the need for fans in the servers.

This strategy reduced overall energy usage, but altered the load distribution within the data center operation. This move had implications for PUE, which compares a facility’s total power usage to the amount of power used by the IT equipment, revealing how much is lost in distribution and conversion.

“If you look at the PUE, the server power went down because I removed the fans,” said Bhandarkar. “And the infrastructure power went up. So if I calculate my PUE, it got worse. But if I look at the overall energy efficiency, of how much work I’m getting done for the total power consumed, I end up ahead.”

Here’s the full video, which runs about 3 minutes.

Microsoft isn’t the only data center operator looking at eliminating fans or reducing their energy use. Yahoo is working with vendors to develop custom racks and server trays that move fans to the back of the rack or cabinet, and provide sleeker trays to improve airflow through the equipment.

Facebook recently worked with its vendors to adjust the fan speed of its servers as part of a broader energy efficiency retrofit of one of its Silicon Valley data centers. “These fans are PWM fans – pulse width modulation,” explained Jay Park, Facebook’s Director of Datacenter Engineering. “They’re typically pre-set by the manufacturer to run at higher speeds. You modulate the fans to a lower speed and you bring less air through the servers.”

At least one server vendor has developed racks that embrace this concept. In 2008 SGI (then known as Rackable) introduced CloudRack, which featured server trays with no covers and no fans, using larger fans in the rear door of the rack to cool the equipment. See our video demo of CloudRack for more.

Get Daily Email News from DCK!
Subscribe now and get our special report, "The World's Most Unique Data Centers."

Enter your email to receive messages about offerings by Penton, its brands, affiliates and/or third-party partners, consistent with Penton's Privacy Policy.

About the Author

Rich Miller is the founder and editor at large of Data Center Knowledge, and has been reporting on the data center sector since 2000. He has tracked the growing impact of high-density computing on the power and cooling of data centers, and the resulting push for improved energy efficiency in these facilities.

Add Your Comments

  • (will not be published)


  1. SeanM

    PWM is described as "pulse with modulation" as opposed to the proper description "Pulse Width Modulation". Assuming it's not a simple typo, it's important to describe the concept: you can say that the fan is fed power for a certain percentage of a second per second (or a certain percentage of a millisecond per millisecond, etc), so it's pulsed on and off quickly. The longer the 'on' period, the more air the fan moves with the fan 'coasting' back down between pulses. It's how you get a gradual increase in effective speed when using a device that only has 2 states: on and off. As an analogy, imagine if your car's engine could only run at one speed, and to change the effective speed to less than full speed the drive wheels were taken off the ground for, say, 1/2 of every second (with the engine capable of running the entire time, but not needing to whenever the wheels are off the ground): you'd get a drive forward each time they were on the ground, and a coast when they were off. If you reduced the amount of time they were on the ground to 1/4 of a second, you'd get less energy pushing the the car forward effectively and a lower average speed, and if you increased it to 3/4 of the time, you'd get more. To get 'max speed' you'd just keep the wheels on the ground and engine running the whole time, and to stop you'd just leave them off the gorund and the engine off.

  2. -BK

    Another great pro for "Cold Aisle Containment". Just need to be sure that all the equipment in the "POD" shares the same demand of heat dissipation requirements. Would like to know what metric and where you are monitoring the temps in the server or the processor, and what are using to control the air delivery; Static or fan speed for cfm?

  3. Nash N

    Good article. I'd point out, however, that the savings, and hence effect on PUE, was, by Microsoft's own admission, rather small. PUE is really meant to look at big changes in efficiency. According to this article the average data center achieves a rating of 1.8 but Google is already hitting an astonishing 1.1.