Green Grid Explains Economizer Findings - By Peter Judge

Views and Opinions on Green IT (Oct 24 2011)

1. 

   Last time I looked at some research the Green Grid did around the use of economizers, which provide “free air” cooling in data centers.

   Economizers are a good thing, because they cool air without using as much power as conventional chillers. Using them should result in lower power bills. It should also give a more efficient data center, lowering the power usage effectiveness (PUE) figure closer to the ideal figure of 1.

   Now, the Grid’s research found that using economizers cut the power usage, right enough.

   The Grid’s survey asked all the data centers what effect the use of economizers had on their data centers’ energy costs and maintenance costs. They reported an average of a 20 percent cut in energy use, and a seven percent cut in maintenance costs.

   The impact on maintenance was less than the effect on energy, and it looks from the results as if a small number reported an increase in energy and/or maintenance costs.

   The next section, however, is headlined “No correlation between PUE and economizer use”, which struck me as odd. If using economizers reduced energy costs, wouldn’t that reduce the PUE?

   I asked the Grid, and got an answer relayed back from the reports authors - it’s all in the way the questions were asked.

   “I think the way we asked the 2 questions leads to confusion of the results,” I was told. “We ask for PUE at one point, then energy savings at another. We didn't ask for PUE before and after economizer use.”

   So, every data center that used economizers was asked whether they had made savings. They were also asked what their PUE was, but not whether that had been reduced by economizer use.

   So if we compare the PUE with and without economizers, we’re comparing between different data centers - and that’s not really worth doing, especially with a small sample.

   Of the 115 data centers polled, 68 gave a PUE figure, and the average of those was 1.69.  That in itself is interesting as a yardstick for how well data center efficiency is going, away from bragging and approximations.

   For some years, no new data center worth its salt has publicised a PUE of more than 1.2, giving the impression that is the norm for newly built facilities, while efficiency gurus have generally said that the average PUE “in the wild” is something like 2.0.  It seems like the actual figure is somewhere between those two extremes.

   The box plots in the paper split the PUE figures between the two and the average (mean) is slightly lower for data centers with economizers, but the difference isn’t statistically significant, the authors explain.

   PUE can vary a lot by climate, and the survey included fifteen from the San Francisco Bay area, but even limiting the plot to these didn’t produce anything significant on a sample that big.

   “The non-correlation between PUE and economizer use speaks to the fact that we did not ask people how they measured PUE, or what PUE was with and without economizer,” the authors told me. “We just don't see a statistical connection between those two values in this data set. If we asked the question differently, we would expect to see the correlation.”

   All clear now? 

   Bookmark or Share this article

Login to comment.