In 2008, the Green Grid organization proposed an industry standard methodology and metric for measuring data center efficiency, the Power Utilization Efficiency factor (PUE). Simply put, PUE measures ratio of the amount of power spent on supporting infrastructure and the power supplied to computing equipment.
PUE = Total Facility Power/IT Equipment Power
The power expended on "IT Equipment Power" is associated with servers, networking gear, and storage. The "Total Facility Power" measures the power spent on supporting infrastructure, such as chillers, air conditioning, batteries, uninterruptible power supplies and fans.
A related metric is the DCiE, which is the inverse of the PUE and represents an efficiency ratio, ranging from 0% to 100%.
A perfect PUE is 1.0, but that would imply perfect UPS systems and the like.
PUE is being popularized in the press currently by many of the big players in industry. The latest and best claim is by Yahoo on June 30 of a PUE of 1.1.
There are many issues to consider when calculating PUE, some theoretical and some practical. For example, in many data centers, incoming power circuits feed both the data center raised floor area as well as support centers, which creates difficulty in parsing power spent on either. Also, how easy is it to measure the power spent on intra-server cooling systems like on-board fans or in-rack coolers?
I had an interesting and humorous observation when I was researching PUE. There is much attention being paid to "free" chilling by using outside air when the ambient air temperature is below the rack intake temperature. This is nothing new! My very first engineering assignment almost 30 years ago was writing a maintenance manual for an air handler that did the very same thing. Only "back then" we used discrete electronics to measure the ambient air temperature and stepper motors to set the correct chiller/outside air mixture. "What goes around, comes around!"
No comments:
Post a Comment