Datacenter efficiency metrics remain a moving target
I received a bunch of emails last week regarding a blog entry I made that indirectly referred to the PUE metric. Most of the negative comments boiled down to either "We know that PUE is flawed" or "Got a better idea?" The Green Grid, the entity that unleashed PUE as a metric, is aware of its shortcomings and is planning on releasing a bunch of refinements by the end of this month.
Technology changes, such as datacenters that generate power on-site, have skewed the metric, since it then becomes possible to have a PUE of less than 1.0, which would indicate more than 100% of the inbound power is being utilized by IT loads. The Green Grid is working on changes to their metrics that will support these types of scenarios. They are also introducing a Power Efficiency Estimator, which is designed as a "what-if?" type calculator that will allow companies that are designing new datacenters or redesigning old ones to look at the impact changes in the proposed topology and equipment will have on the efficiency of the final product.
They've also released an online service, the Free-Cooling Estimated Savings calculator that allows the user to estimate the dollar value of energy savings that can be achieved by utilizing external airflow, which can be of huge benefit in cooler climes. And don't underestimate the value of this "free" cooling; Iceland (yes, the country) has been actively recruiting datacenter investment with its promise of free cooling due to the environment and cheap (and exceptionally "green") geothermal power availability (Iceland produces more power than they can currently consume domestically).
Back in January the country was able to announce that their first major data center investment had been made, with plans to convert the former NATO command center in Keflavik into a data center campus. Apparently investors are unconcerned about the county's predilection for earthquakes and volcanoes and believe that the disaster recovery investment necessary for the data center will be more than offset by the dollar savings in energy expenditures. I've also yet to see a client list for the projected datacenter campus but I'll be on the lookout for it.
On the other side of the world, Australia already has building energy efficiency metrics in place via NABERS (National Australian Built Environment Rating System) which takes into account the entire energy profile of a building, meaning that a building which contains even an efficient datacenter (from the PUE perspective) can still achieve a horrible NABERS rating. The Aussie government is aware of this and is launching their own investigation into designing proper metrics that take into account more than the simplified model of PUE. Of course, we then need to consider if a government-based metric have more of an impact that one that evolves in the industry, or does it end up being a government-mandated requirement for datacenters in Australia, forcing vendors to take it into consideration worldwide?
And moving back to the other side of the world again, exactly what metric will take into account the energy efficiency of the data centers being implemented under Helsinki, Finland, where the cooling is provided by water pumped in from the Baltic (with an average temperature of only 46 degrees F for 7 months of the year) and out to the central heating system used by the city which delivers water for heating directly to thousands of residences. This means, that unlike most data centers, the waste heat has actual value, since the city needs to heat the water it supplies for heating anyway, and having it delivered at a higher temperature than would otherwise be found saves them money.
There will never be a "one size fits all" metric for data center design, but whatever metrics do evolve will be useful to the people whose jobs are designing and implementing energy efficient data centers.