Jim Smith, Chief Technology Officer, Digital Realty Trust, addresses the crowd at 7X24 Exchange on areas of incremental efficiencies to be examined in the data center. (Photo by Colleen Miller.)
– The data center industry’s push for ever-better energy efficiency has become a tale of two PUEs. There’s the exceptional Power Usage Effectiveness (PUE) scores being reported by companies like Google
that design their own custom servers. And then there are the slightly higher PUEs available to the best multi-tenant data centers and enterprise customers with complex infrastructure.
Few understand this better than Jim Smith, the Chief Technology Officer of Digital Realty Trust
, the world’s largest operator of data centers. Smith is passionate about energy efficiency and data center design, and well-versed in best practices. But he also knows that not every data center can emulate the strategies used by Facebook and Google.
“There’s been a bifurcation of PUEs and expectations,” said Smith, who spoke yesterday at the 7×24 Exchange Fall Conference
in Phoenix. “We see quite a bit of confusion in the marketplace. We see PUEs of 1.05 and 1.1.”
Single Tenant versus Multi-Tenant
But when those ultra-efficient scores earn publicity, some may assume such scores can be achieved by any data center provider. In reality, Smith says, PUE potential is linked to the facility’s purpose and users.
A single-tenant data center – like those operated by industry titans Google, Facebook and Microsoft – can be optimized for custom hardware. As a builder of multi-tenant wholesale data centers, Digital Realty must build a design that can work with a broad range of hardware and customers.
“We can’t do what Facebook does, because we don’t control the box (server),” said Smith. “People are taking the box, and building a facility around it. We can’t do that, because we see a more traditional compute package. I can’t take a data center like (Facebook’s Prineville facility) and put it in front of most of my customers. We always have to design for the most particular customer, which is usually financial services.”
With more than 110 facilities spanning 21 million square feet of space, Digital Realty has extensive experience developing data center space. It builds data centers for other providers, including Equinix, IBM, Terremark and SoftLayer, who are among Digital’s 570 customers.
Digital Realty has been active in evaluating the latest efficiency trends. The company has developed designs that integrate some elements of Facebook’s Open Compute Project, and is using pre-fabricated modular components to accelerate deployment of its electrical rooms. The efficiency gains have had a significant payoff, reducing the company’s power bills across its data center footprint.
Nearing Point of Diminishing Returns
So where does Smith see the frontier for multi-tenant PUEs?
Smith cited Digital Realty’s new data center in Sydney as the state-of-the-art in energy efficiency for multi-tenant buildings. The facility features many traits of a traditional enterprise data center – a raised floor, 2N redundancy, and a double-conversion UPS system, widespread use of variable frequency drives (VFDs) – but has a PUE as low as 1.18 when using fresh air cooling.
After working for years to refine Digital Realty’s data center design, Smith says it will be difficult to improve on the PUEs in Sydney.
“I’m approaching my limits,” said Smith. “We find ourselves at a point of diminishing returns.”
Smith isn’t alone in this dilemma. As the Googles and Facebooks of the world push for ever-lower PUEs, they are also assessing the costs and benefits of the strategies that could take a facility with a 1.10 PUE and move it closer to the magic 1.0, the best score achievable. In some cases, techniques that would improve PUE don’t make economic sense. In other scenarios, they involve tradeoffs that could impact reliability.
The PUE metric championed by The Green Grid
compares a facility’s total power usage to the amount of power used by the IT equipment, revealing how much is lost in distribution and conversion. These numbers demonstrate the potential for ultra-efficient data centers, but are sometimes critiqued as having limited relevance for enterprise data centers that must engineer their facilities for many types of applications and workloads.
To be sure, many older facilities and corporate data centers can see enormous benefits from improving their PUE. The average Power Usage Effectiveness (PUE) rating for data centers is 1.8, according to a survey of more than 500 data centers conducted by The Uptime Institute. For companies at a 2.0 PUE, improving to 1.6 or 1.5 can produce major benefits.
Geography is one of the areas where strategies are diverging. Single-tenant server farms are being built in places like Prineville, Oregon or Quincy, Washington 
or western North Carolina 
. These locations all offer abundant land for massive facilities that can house of tens of thousand of servers, and the electricity to power these armadas of servers.
But that isn’t an option for many multi-tenant service providers. “It’s clear that most data center customers don’t want to be in remote places,” said Smith. “Many, many, many customers want to be close to their equipment. Our strongest market is Silicon Valley.”
What about new approaches to power sourcing? Smith is intrigued by the potential for fuel cells, but cost has been a historic challenge. That’s beginning to change.
“It’s the first time in all my years in this business that the economics of fuel cells have started to make sense,” he said. “Even though the economics are favorable, we’re having a big, robust discussion among our management (on the topic).”
Article printed from Data Center Knowledge: http://www.datacenterknowledge.com
URL to article: http://www.datacenterknowledge.com/archives/2012/11/13/are-low-pue-ratings-hitting-their-limit/
URLs in this post:
 Quincy, Washington: http://www.datacenterknowledge.com/archives/category/regional-markets/quincy-wash/
 North Carolina: http://www.datacenterknowledge.com/archives/category/regional-markets/research-triangle/
 Rich Miller: http://www.datacenterknowledge.com/archives/author/richm/
Click here to print.