PUE : details and controversy(ies)

Following previous article, I found this interesting piece from Datacenter Knowledge and so decided to vampire it and dig more into related contents.

First, this nice schema well introduces PUE:

As more and more PUE adopters are releasing into the wild PUE values (improbably low numbers, lower than competitor if possible), a controversy is starting on the reality and seriousness of theses figures. PUE is simple to calculate with only basic guidance, evolving to an industry standard without the discussions and collaboration typical to norms. Fast adoption could be an evidence that it is a too easy game:

• approximations (measurement, weighted average) : It is complicate to measure everything in a datacenter, like it should be for PUE. Old infrastructure are not that flexible and when it is not designed from begining, it can’t evolve. If your datacenter is in an office building for exemple, it could be that generator, power conversion, UPS and cooling  are shared amongst many usages and users.  Google for exemple is not taking into accounts small values so small datacenters.  The sampling method is also important in any measurements and as coolings is highly linked to seasons, should not be discarded in reports. PUE varies! See illustration from Microsoft who has a hidden modesty showing it.
  
• use of products’ manufacturer technical value instead of real one. It can’t believe they went to every server, every PSU to measure it. They just make maths from the spec sheet. I do. I wish I could measure it, some servers give that information but it is just easier to get to the point. And more scalable.
• restricted scope : PUE is often calculated for modern,  self-owned facilities ? How about the one you lease, the racks provided by your telco to host internet services, etc. PUE is used then as a justification for investment in new and shiny datacenters, not as a metric for IT efficiency.

Again, PUE is not a norm, standard or certification, it is just a metric.

It remains the way to go of course, incorporated in a global effort, involving desktops in the scope…

It is always good to pay attention to this.

Facts and links after the jump.

To sum up and start from Datacenter Knowledge article :

So VIP users of PUE (PDF) are:

• Google (1.21 in limited scope – 6 datacenters, many more known)
• Microsoft (New datacenter container : 1.22, target 1.125 PUE by 2012 thanks to 4th container datacenter generation)
  • MSDN Power of Software article
  • Digital Cave : Data Center Energy Efficiency White Papers produced by Microsoft.
• Sun Microsystems (1.28 at SCA11-1500 Santa Clara datacenter)
• HP (1.25 for POD container)
  
• Digital Realty Trust.

Google controversy at Tech hermit

• Google shoots but misses PUE Target
• Google clarifies PUE Post critiques

Uptime Institute (tier efficiency of datacenter promoter) vs The Green Grid:

• Uptime Companies Gaming PUE Numbers
• Uptime warns data center pros against being benchmarked on PUE
• Uptime is developing its own metric and this will be presented in their Symposium 2009, four metrics proposed by Uptime.
  
• Microsoft and Google as datacenter enemy
  • In disapointment, there is opportunitye) (by a Microsoft employe Michael Manos)
  • Uptime Institute to IT, Microsoft and Google are your enemy by Tech Hermit
• Uptime Warns Data Center Pros on Using PUE, a Simple fix

Related Posts:

• The Uptime Institute’s IT Symposium 2009 non attendance report
• The Green Grid Technical Forum
• Week 51 Links
• The Green Grid European Free Cooling Tool
• How a Good Metric Could Drive Bad Behaviors