What's in a number? PUE/DCIE

So Google recently published their quarterly PUE numbers for the data centres they are publicly reporting on. I've met a number of guys from Google and guess what, they're no pushovers when it comes to understanding what value people get out of PUE.

I refer back to late summer 2008 when I was with a small group of industry reps in Ispra at a close meeting of the European Code of Conduct for data centres. Christian Belady was there having not long joined Microsoft and Liam Newcombe was there also, they were violently agreeing that PUE/DCIE was a call to measure.

There was even talk about T-shirts being made with 'Stop talking - Start measuring!' printed on them. I even wrote a recent blog asking if these simple data centre efficiency metrics put out there by the Green Grid had reached their expiry date or not.

As Liam would tell you (and he often does to unsuspecting listeners), metrics are either simple and of minimal value or complex and of high value, it is a trade off, life is full of them!

As the high-tech industry that we are and for all the complexity that the world of computing deals with, we still love to boil things down to a single dimension, a single number, not even a series of numbers! I guess it's hard for our busy brains to deal with anything more.

However there is life in the old (ok....still rather new) metrics (PUE/DCIE) as far as I see it and it is also possible to do the impossible and make a human understand more than a single number. The trick is to present it visually and not just as a series of numbers or data points but as a shape. Human brains are better at shapes and patterns generally.

So, let's have an example. If I told you my data centre had a PUE of 2.0, you could draw a number of conclusions from that. You could say that for every watt consumed by IT equipment my mechanical and electrical infrastructure used (or wasted) another watt keeping the IT equipment cool and power protected, etc.

As a data centre manager however you might ask, well is that an average PUE? and if so what time period was it taken over, oh and by the way was the IT load static during that time or did more equipment get installed or perhaps decommissioned? What time of year was it and where abouts in the world was that data centre (i.e. what was the external climate like).

And this is the issue! Once we get beyond the initial value of such a simple metric the follow-on is badly lacking.

Now take a look at the following graph.

What is it telling you? Well lets look.

The vertical axis is DCIE. The left axis is external temperature (in Celsius) and the right axis is how full the data centre is (or how much of my provisioned IT capacity I'm currently using).

Now what is the surface plot showing me? It is showing me what happens to my data centre efficiency at any external temperature and IT load point. The shape itself is interesting though as I can now see how my data centre performs under almost any combination of IT load and external conditions.

There is also an interesting slope in the surface which is telling me this is an economised data centre. I can also see that the economised has a more noticeable effect the closer my data centre gets to being full.

That translates to maximum efficiency being achieved when I can closely match my provisioned capacity with the actual drawn capacity. Excess, unnecessary, underutilised or unused capacity is the enemy of efficiency!

Now imagine I had a DCIE plot like this for all the data centres I manage. I can visually see how each one performs, what they are sensitive to in terms of external temperature and load. A simple metric made more useful.

NOTE: The DCIE surface plots were generated by the Romonet data centre simulator.