PUE is calculated by dividing the amount of power entering a data centre by the amount of delivered to the IT and communications equipment. It therefore reflects losses that occur during power conditioning, etc, and the power needed to cool the equipment.
The closer the PUE is to 1, the greater the efficiency.
John Dwyer, area data centre manager - international at Microsoft, told iTWire that Microsoft's PUE was 2.0 in 2005, but it has now been reduced to 1.6. This was achieved in part thanks to the company's Dublin (Ireland) data centre, which takes advantage of the mild climate to use fresh air cooling.
Doing away with traditional cooling systems saves power and also reduces water consumption. For comparison, the Microsoft data centre in San Antonio (US) uses 1.8 million litres of recycled water per month.
So what is the Dublin PUE? Find out on page 2.
Given that a PUE cannot less than one, achieving Microsoft's self-imposed target of an average PUE of 1.125 across its data centres requires "dramatic" changes, he said.
One way this will be achieved is by moving away from traditional concepts (eg, duplicated power supplies, massive banks of standby batteries) used to provide high levels of availability through hardware, instead relying on geo-diversification.
Rather than insisting that any particular data centre should stay on the air in the event of power outages and other problems, a cloud infrastructure merely requires that the services (including the associated data) are always available from a data centre somewhere in the world.
"It's an evolution," said Dwyer, "we already have a geo-diverse strategy" that can withstand the catastrophic loss of a location.
Microsoft is also adopting the modular, containerised approach to data centres, packaging as many as 2500 servers into a single 40ft container along with supporting services.
How does this help improve the PUE? Please read on.
It also makes it easier to dynamically populate data centres with just the right amount of IT equipment that's configured to suit particular purposes. Over time, software will provide more of the resilience required (rather than making the hardware do all the work) and it will be a simple matter to switch out containers for others with lower configurations (eg, no UPS) but improved efficiency.
Furthermore, the use of standardised modules makes it possible to provision centres more quickly to support the delivery of new services.
While Dwyer (who is is responsible for Microsoft's data centres outside the US, including those in Amsterdam, Singapore, Hong Kong and Tokyo as well as Dublin) doesn't expect Microsoft will offer data centre design services, he did say "we will... continue to share our experiences with customers" as well as working with OEMs to help evolve data centre products.