I'll bet you a FVER

 

I've spent close to three years now as a member of the Global Metrics Harmonisation Task Force, an informal gathering of representatives from the EU Code of Conduct for Data Centers, the US-EPA, US-DoE, the Japanese Data Center Council and the Japanese Ministry of Economy Trade and Industry.

Like many other bodies, formal and informal, one of the many metrics we've been searching has to do with something the industry generally refers to as "Productivity". Generally this is refered to as the amount of useful work I achieve for a given unit of energy consumption. Productivity was seen as a logical next step to the very successful PUE (Power Utilization Effectiveness) metric in trying to understand what I achieve for all the energy my data center consumed.

Indeed, it is a very sensible question to ask at first sight and one that seems has a lot of value should I be able to compute such a metric. In fact, the Green Grid developed a so-called mathematically correct productivity metric called DCeP or Data Center Energy Productivity. Like many of the other approaches it was hard to understand, hard to implement and contained a lot of 'factors' that could, if tuned, give vastly different results.

Now don't get me wrong, I'm not trying to poke holes, not at DCeP, the productivity proxies (although there are a few here that I have poked large holes in), the Japanese DPPE metric or any of the other attempts to try and measure productivity per unit energy consumed. I'm simply want to point out that none have really gained much traction, unlike our old friend PUE. Why?

Metrics like many things in the engineering world are a trade off, there is no such thing as a perfect metric. There is certainly no such thing as a metric that cannot be gamed in some way by those that wish to misuse them.

Liam Newcombe, our CTO, who also is the secretary of the BCS Data Center Specialist Group, came up with what I believe is a simply briliant idea of turning the whole issue on its head. What is the perfect proxy? How can I compare a video streaming website with a super-compute facility as far as productivity is concerned. What is useful work anyway? Who exactly is it useful to?

Liam pointed out that as there is no one answer for which proxy to use, why not just let people use the one or ones that suit their business best. Ah! I hear you say, but how can I compare my performance with someone else? How can I have a simple metric like PUE that with one number will give me a great indicator if I'm doing well compared to others, or compared to myself over time?

That's the second part of Liam's idea. Measure the waste rather than the useful work. We have trouble defining work because it's different for everyone and what units do they have anyway! The waste however is rather more obvious; if we are talking about energy productivity then the waste is energy, measured in kWh.

The other smart element of the metric Liam developed recognised that the issue in the case of IT is not that it is the energy efficiency of devices themselves, but rather their very poor utilisation and high levels of fixed energy overheads. There are actually some very old white papers of Liam's that discuss fixed vs variable energy within a data center; however he's now brought all this together into a metric called FVER - Fixed to Variable Energy Ratio that neatly sidesteps the big issues that most productivity metrics have and is simple enough to implement with a clipboard and pen, and the answer is also one number just like PUE.

FVER is by no means perfect, I've already said it sidesteps a bunch of issues, but it's simple enough, it goes beyond PUE and focuses people on the waste, something we all want to aim for 0% ideally.

Have a read of the presentation and white paper and give it a try today!