Time's up for PUE/DCIE?

When the Green Grid first announced its data centre efficiency metric PUE (Power Usage Effectiveness) there were cries from some about it being a 'sleight of hand' trick by vendors to focus pressure on the inefficiency of the mechanical and electrical plant in a data centre rather than the IT equipment they house.

Now of course the real reasons why PUE has been successful are; that it is simple to measure and simple to report and these two characteristics have gone a long way to making up for the many shortcomings of PUE. Also, it focuses people on at least part of the data centre efficiency problem. Sure, server efficiency, utilisation and software efficiency are probably the big win, but starting at the source of the energy chain makes some sense.

Little did the Green Grid know that suddenly people would jump on the bandwagon and start using PUE as a marketing statement about their data centre. It was not even the claims of PUEs less than one that worried many but the fact that as is inevitable when any ‘scoring system’ is put into place, people started to compare their PUE numbers between each other.

Almost two years ago we published via the BCS Data Centre Specialist Group a paper on data centre efficiency metrics which explained many of these shortcomings and advised that using PUE / DCIE to compare data centres made little sense and that using PUE as a target would be counterproductive.  Earlier this year the Green Grid also announced that PUE shouldn't be used to compare different facilities as this could lead to misleading conclusions.

As many have said before me 'stop talking and start measuring'! The metric itself is less important than the act of measuring and then asking why. PUE is clearly a good place to start but a closer look at what you measure and how you should use and interpret this metric is important.

I'm pleased to say that I talk to more and more people who complain that a single number cannot represent the complexity of their data centre and its efficiency. When I show them one of our DCIE (or PUE) surface plots they generally light up and say 'That's exactly what I've had in my head but don't know how to produce it!'

This is the advantage of simulation. With simulation I can run my facility through its entire capacity range and also at different external temperatures and generate a three-dimensional surface that allows me to properly understand the efficiency of my data centre across its entire operating range.
I hope one day to be presented with such graphs in every data centre I visit, meaning that those managing the facility have evolved their understanding to a point that makes PUE a truly valuable metric.

So in summary, time is most certainly NOT up for PUE/DCIE, particularly as there is no sign of any useful IT efficiency metric but we need to evolve the way we take and analyse our measurements into something more useful for actually running a data centre and making decisions.

NOTE: The data centre simulator (DCSIM) can produce a PUE or DCIE surface plots (as per the image at the top of this article) based on an understanding of the data centre M&E layout and the energy performance of each node as 'estimated' by the manufacturer