Gaming the PUE: Will Cooling Emerge As the Big Green Data Center Investment?

We've been investigating exactly how the new ENERGY STAR for Data Centers works and sharing what we've learned. A look the first ENERGY STAR data centers reveals an interesting pattern: a foucs on reducing energy by tackling cooling, not ICT equipment and use.

We've noted that ENERGY STAR's rating system relies on PUE, the ratio of total ICT facility energy use to ICT gear energy use. So reducing the non-ICT-gear energy consumption lowers (improves) the PUE while reducing the energy consumption of the gear increases it. And since cooling is by far the largest component of the non-ICT-gear consumption, that's a logical place to leverage a better PUE.

A look at how the first three ENERGY STAR data centers earned their ratings reveals an almost exclusive focus on cooling. No mention of ICT energy-saving tactics like of equipment consolidation, server virtualization, data de-duplication and archiving, server power management, etc. Why? Because while these would contribute to even lower total energy consumption, they would also decrease the denominator on the PUE equation. That would definitely increase the PUE and might potentially jeopardize ENERGY STAR certification.

It is not surprising, given this artifact of PUE, that a projection of the global green data center market forecasts "power and cooling infrastructure solutions will be the largest portion of the green data center market opportunity".

Case A
Assume a data center with a PUE of 2.0 and with cooling consuming 72% of the non-ICT-gear energy.
PUE = ( ICT-gear energy + cooling energy + other non-ICT-gear energy ) / ICT-gear energy
PUE = ( 1 + .72 + .28 ) / 1 = 2 / 1 = 2.0

Case B
A 50% reduction in cooling energy consumption (36% of non-ICT-gear consumption) would yield:
B. PUE = ( 1 + .36 + .28 ) / 1 = 1.64 / 1 = 1.64
This is a significant PUE improvement and an 18% reduction (from 2.0 to 1.64) in total energy consumption.

Case C
Now, let's add a 36% reduction in the energy consumed by the ICT gear, which we'll assume reduces cooling 36% from Case B.
PUE = ( .64 + .23 + .28 ) / .64 = 1.15 / .64 = 1.80

See the problem? Total facility energy consumption dropped from B to C, but PUE turned back up. There goes PUE bragging rights, but what about the ENERGY STAR rating? We've noted the complexity of the relationship between PUE and ENERGY STAR and, according to our team's expert, this increase in PUE would not necessarily lower the ENERGY STAR rating. But it does appear that one can get a better ENERGY STAR rating by reducing your total energy through non IT energy reduction than through and equivalent ICT energy reduction.

We trust that the value of reducing total energy consumption and peak power demand will help data centers avoid the temptation to game PUE by limiting the implementation of virtualization and related tactics once cooling has been addressed. Work on ICT efficiency would still pay-off in avoiding UPS and rackspace investments as the workload increases, as well as future utility power constraints.