The rise and fall of PUE - By Peter Judge

Views and Opinions on Green IT (Apr 12 2010)

1. 

   In one sense PUE is declining: every data center in the world is watching its "Power Usage Effectiveness", and trying to get it down because a low PUE means a more efficient data center, with more of its energy going to the servers, and less to the air conditioning. Historically Average PUE has been two or more, and now sites are pushing towards 1.2 or thereabouts. 

   But in another sense, PUE is definitely on the rise. It has been adopted as the international measurement for energy efficiency in data centers, by the US Department of Energy, the European Union and the Japanese Ministry of Economy, Trade and Industry. So it looks like becoming the yardstick by which data center people's work is judged - the "miles-per-gallon" of the industry, if you will. 

   
   There is a lot to be said for a measurement that is simple enough to be readily applied and understood, and that is the basis on which PUE is being put forward, originally by the Green Grid, and now by the whole industry. Plenty of people have pointed out difficulties with the measurement, including people within the Green Grid itself. 
   
   

   Some have said that PUE takes too limited a view of what goes on in the whole lifecycle of a piece of IT equipment - so chasing a lower PUE might mean that kit gets replaced before the end of its useful life, saving the data center's energy costs, but at a net cost to the world, because of the unaccounted costs in the rest of the lifecycle. 

   Others have warned that PUE is too blunt an instrument, and may alter behaviour artificially and in unintended ways, because of limits to its definition. For instance, raising the temperature in a data center might lower the PUE, because the air conditioning can be switched on less. But it's perfectly possible that in the new settings, the same energy or even more might be used to cool the servers, and the center might be no more efficient. 

   Raising the temperature of a rack of servers might simply have the effect of causing the internal fans to operate more. Because the fans are inside the server, their energy is counted as going to the IT system, and reduces the PUE - even though a rackful of little fans might turn out to be less efficient than a big aircon unit. 

   More subtly, some have warned against what will inevitably happen if efficiency is boiled down to one number: it will be used to compare data centers. “PUE is a relational metric,” said Vic Smith, EMEA data centre strategist at Dell, and a senior member of the Green Grid. “It’s not meant to be used as a comparison between data centres”. 

   On some basic levels PUE can't be used for comparison, because it depends on so many things that are outside the control of the data center manager. Sites in cold countries have the benefit of free air cooling, so need to use less energy (and on this site, I feel we could mention Iceland). Good luck to those sites, and if all other factors are taken into account it seems a good idea to push as much as possible of the world's cloud computing load towards data centers with a low PUE. 

   Going up a level PUE doesn't look at the work output of the sites. Centers where there is a high utilisation of servers and disks might generate more heat, and more power to cool the systems down, winning a poor PUE score, even if overall the amount of energy used per MIP of processing might be lower. 

   And going higher still, how useful are those MIPs? Suppose a site is well tuned and using a dribble of energy for every MIP of processing power compared with others, but that processing effort is being used to drive horrendously inefficient processes, with massive duplication of effort. PUE is never going to spot that one!

   Someone once said: "For every problem, there is a simple, beautiful and elegant solution...  Unfortunately, that solution is wrong." 

   That's probably as true of PUE, as it is of everything else. But even with all that taken into account, I think PUE takes a good stab at measuring the problem, and gives us all something to discuss. 

   As long as we're careful not to rely blindly on it, or use it inappropriately.  

   Bookmark or Share this article

Login to comment.