Is PUE the new battleground?

I’m reading a lot about data centers of late – so much so that I’m even spelling it the US way already (sigh). What is becoming increasingly clear to me though is that PUE may well be the new battleground between some of the industry heavyweights.

In very basic terms, PUE (Power Usage Effectiveness) is the ratio of power incoming to a data center to power used. The theoretical ideal is 1.0 of course which means you’re not wasting any energy. As Mike Manos points out this is all part of the “Industrialization of IT” that he and our GFS team works on. Google and Sun have both been waxing lyrical about PUE of late with some impressive numbers, particularly from Google who cite a PUE of 1.13 in one datacenter. Very impressive indeed.

This led me to look at the table in Manos’ post a little more closely where I noticed a PUE of 1.11 for our shiny new Chicago facility but that goes up to 1.22 when you add mechanical.


As Mike points out though

For the purists and math-heads out there, Microsoft includes house lighting and office loads in our PUE calculation. They are required to run the facility so we count them as overhead.


My takeaway is that PUE is something that customers, regulators and environmentalists are going to start looking at with much more scrutiny over the coming years. Don't’ be surprised to seeing this stuff show up on earnings reports and annual reports. For it to have any real value though, we need to get to some sort of consistent measurement.

Comments (4)
  1. Steve,

    I would suggest that PUE is the current battleground…..but it is only the first battle of the very long and bloody war.

    Given some of the shortcoming of PUE (which I am sure you are aware of) I expect further work to take place developing something that is as easily understood but that takes the efficiency/effectiveness of the compute platform into account.

    Until then….I guess we will have the line "My PUE is bigger than yours !!!!"

  2. tinkthank says:

    Both PUE and DCiE are gamable metrics.  At the end of the day the efficiency of the datacenter matters less than the amount of energy consumed to perform a set amount of work.  Allowing servers to idle eight or more hours during lulls in traffic even at 1.2 PUE is not an effective or efficient use of energy.  Worse, shifting facility load (air movement) from the facility into the server makes PUE metrics look even better.  I can name two internet properties doing this.

    The biggest issue may be that PUE is referenced to server power draw.  I’ve tested many 8 core Xeon servers and seen power draw range from 250 up to 350 watts for the exact same config depending on server vendor.  So what’s better?  A PUE at 1.2 with 350 watt servers or a PUE of 1.3 with 250 watt servers?

  3. Gareth Brown says:

    Another measure is the data centre’s infrastructure efficiency, which is 1 / PUE x 100%.

    PUE is a good metric to use across organisations who require a benchmark across their own data centres (or computer rooms if you have a measurable feed)- The accepted rule is straightforward and offers a measure of efficient equipment AND best practices.

  4. tinkthank says:

    No, PUE and DCiE do not provide a straightforward way to measure equipment efficiency.  They help gauge the efficiency of the physical plant, but not the server.  And it fails to capture the interaction between servers and the physical plant.

Comments are closed.

Skip to main content