Metrics for Australian Data Centres

The first talk I attended at Data Centre GreenTech Melbourne 2010 was William Ehmcke and Graeme Philipson from Connection Research (to be renamed "envirAbility") explained they have a contract with the Environment Department to produce data centre metrics for Australia (for release later in the year). With all of this it may be time for an update to my book "Green Technology Strategies" and the e-learning courses for Open Universities Australia, Australian Computer Society, and the Australian National University.

Graeme argues that the deficiencies of current metrics, such as Power Usage eEfectiveness (PUE) and Data Center Infrastructure Efficiency (DCiE) are overstated. Current measures of efficiency are limited to the efficiency of the data centre to deliver power to the IT equipment. The idea is to measure how much energy wasted on cooling and the like. Not all the energy delivered to the IT equipment is usefully employed, but measuring how much does useful work, is difficult.

Ideally there would be a measure of Data Centre Energy Productivity (DCeP). This would be the ratio of "useful work" to energy input to the data centre. However, measuring what is "useful work" is not simple. Efficiency could vary depending on what computer equipment is installed, what software is used and how busy the system is. Because of the difficulty of measuring "useful work", industry groups typically set an arbitrary figure of 5% (that is 95% of the energy is wasted in a computer).

US EPA are to issue PUE based metric next month for energy efficiency. As EPA measures are commonly used, this new measure is likely to be widely adopted. This will be a 1 to 100 scale. It should be noted the PUE measures the efficiency of the data centre equipment, not the computers in it.

They mentioned that CompTia sustainability practices examination to be released in the next month, which includes their framework. However, CompTia were previously planing to have this out in December 2009.

The Green IT Promotion Council (GiPC) of Japan (グリーンIT推進協議会), have developed a draft Datacentre Performance Per Energy (DPPE) measure. Unfortunately the material has not yet been translated to English. I was able to use a machine translation to find:


DPPE is a combination of the following four elements ...

• DUE (Data Center Use Efficiency): Effective Use 電力効率 Power Efficiency
• ITPE (IT Performance per Energy): processing of
  能を電力で割った値 Divided by the power capacity
• D C i E (Data Center Infrastructure Efficiency) の逆数でエネルギー効率指標 PUE energy efficiency indicators in the inverse
• GPE (Green Power Efficiency): Natural Energy Utilization

From: Green Data Centre Trends, KIIS Quarterly, Volume 6-1, December 2009 (データセンターのグリーン化の動向 )

Australia has the NABERS environmental rating for office buildings. There will be mandatory disclosure of this rating for building sales and eases for larger buildings from July 2010. Because data centres consume so much energy they were distorting the building ratings. The example was given of the ABS Building in Canberra (one I know well) which had difficulty with its energy rating due to the large data centre.So Envirbility are developing a NABERS rating for Data Centres for the Environment Department. This is due to be completed in four months time. There is no mandatory disclosure for data centre efficiency currently scheduled. One interesting comment was that because these data centres are within general purpose buildings, there is less variation in efficiencies depending on location, than for dedicated centres.

Envirbility argue that PUE is a reasonable measure. Others such as Compute Units Per Second (CUPS) from Emerson, places emphasis on the processors, not other equipment. It should be noted these measures would be useful for comparing similar equipment and applications only.

There were interesting questions about how holistic measures should be. As an example, there are power losses between a remote power station and a data centre. If this is not included it will penalise co-generation plants, which do not have as high transmission losses.

As an applications software person, I find these measures of efficiency less than encompassing. There seems to be little point in saving a few percent on power efficiency if this is used to power a web server which is delivering documents which are so poorly designed they are wasting 99% of the capacity.