Workload distribution isn’t really new.  But distribution based upon facilities needs (like reliability / efficiency) as in this example at Google is new (at least to me  Frankly facilities-smarts  seem a requirement to ensure moving workloads for other purposes – preserving performance under load, for example – doesn’t yield unintended and disasterous consequences.  Imagine if turning on VMWare Vmotion moved workloads to servers that looked highly available from a performance point of view, but were dangerously close to overheating  because they were in a hot zone.

But beyond risk avoidance, factoring facilities concerns like efficiency into workload distrution has major upside in improved efficiency.  And this is something that virtually all data center professionals would benefit from as it doesn’t require changes to the physical plant.  Better still, the efficiency gains are greater as every unit of energy saved on the IT side saves another unit on the facilities side through cooling avoidance.

I’d say a 2x improvement that doesn’t require major facilities changes is pretty cool.

If you can work your way around the pot shots at marketing his latest post is no exception as it details some shortcomings of the Green Grid’s Power Utilization Effectiveness ratio.  Through a new construction and roll-out scenario, he contrasts the difference between design target PUE and various forms of assessed PUE (like annual average and annual peak), crisply demonstrating how the numbers dramatically change over time, despite generally adhering to best practices.

As Mike says, the actual PUE rating doesn’t really mean anything; it’s the discussion of what’s behind the number and how it’s changing that’s important.

The Data Center Efficiency Lifecycle

In addition to effective and simple metrics (PUE is certainly a start) the industry needs  consistent and regular measurement.  Today this practice is actually far harder to find in the wild than self-proclaimed, seemingly low PUE ratings. As I’m sure others have pointed out (please let me know who), data center efficiency is a journey not a destination.  To guide our way, we need to define a high-level process and standardized data collection which provides not only credible point-in-time metrics, but allows us to track our progress over time.  Only then can we not only consistently assess our efficiency, but also document the effectiveness (or lack thereof) of various transformations.   To this end, I propose the Data Center Efficiency Lifecycle.  Far from rocket science, I know  (I modified it from the widely used process which enterprises use to manage IT security.)  Feel free to edit the content of the steps to suit your needs, or to add additional steps that would tune it to your environment.  As with PUE itself, the point is not that the exact same process apply to everyone, but simply that we have a continuous process to foster regular improvement, drive a dialog across the organization, document progress over time, and (perhaps) support regulatory / compliance requirements.

What do you think?  Are you already doing something like this at your organization?

Details continue to stream in about Yahoo’s planned datacenter outside Buffalo, New York.  This time regarding their planned “chicken coop” style design which includes louvers to maximize the use of outside air for cooling.

Leveraging this and other efficient design techniques, they are aiming for a PUE of just 1.1.  That’s impressive.

Our sources indicate that Yahoo will use a new line of energy efficient servers from HP dubbed the Red Rooster, which run on – you guessed it – chicken feed

If only it were that easy.

Well done Yahoo.  There is no doubt plenty to do bringing this dream to fruition.  Thanks for taking the time to share your innovation with the rest of us.