Google Data Centers More Efficient Than the Industry Average

16 thoughts on “Google Data Centers More Efficient Than the Industry Average”

  1. Some of Google’s numbers are suspect. While many are believable, their data for facility “B” is outside the envelope of current engineering. And while their search algorithms are wonderful, their mechanical engineers are just like the rest of us. PUEs of 1.2 in a very large facility with lots of sensors, in the right locations: possible. PUE’s of 1.15 or below? Not with our current cooling technology.

  2. Well if you have a data center in Alaska with excellent air-flow design, I suspect you don’t need to spend anything on additional cooling other than server fans.

    It’s a fascinating read, which unfortunately skimmed a lot of details.

  3. I’m excited to see Google take such a strong approach to promoting greater data center efficiency. But, I think what’s lost in this discussion is one of the most fundamental issues with the efficiency of a data center: power distribution. Google does call it out on their website(http://www.google.com/corporate/datacenters/step1.html), saying, “Up to a third of the total energy consumed by a typical server is wasted before reaching the computing components. The majority of these losses occur when converting electricity from one kind to another. The power supply, which converts the AC voltage coming from a standard outlet to a set of low DC voltages, is where most of the energy is lost.” While technologies like “evaporative cooling techniques” are certainly important in taking excessive heat, by fixing the problem at the root – inefficient facility-level electric distribution – you can generate less heat in the first place (which means less to cool). This occurs because, as Google notes, there are conversions (AC to DC), as well as transformations (higher voltage to lower voltage), that need to take place in a data center, which create excess heat at each point of change. With AC power distribution you are looking at 5-7 conversions and transformations, opposed to 2 with DC power distribution – reducing energy consumption anywhere from 15-50%. When it comes to energy efficiency in the data center, I’m as big a fan as anyone of unique and inventive technologies, but if you simply start at the source (electricity), you may be surprised how far you’ll get (www.validusdc.com)

  4. I fear that this post is more aimed at marketing material than real science. The math simply doesnt add up and their definition of PUE is highly circumspect and they dont event use the conventional use of the metric instead opting to create their own version of it which excludes components. I applaud their effort, but its obviously gamed. They dont even include smaller facilities which would naturally raise their average numbers. I have a more detailed post at:

    http://techhermit.wordpress.com/2008/10/02/google-shoots-but-misses-pue-target/

This site uses Akismet to reduce spam. Learn how your comment data is processed.