This audio was created using Microsoft Azure Speech Services
As I get older, I have found I more easily get frustrated. This is not good news since I want to avoid turning into a grumpy(er) old(er) man. A recent irritation is the number of questions I need to ask to truly understand a data center’s PUE (Power Usage Effectiveness). Here’s the scenario: A newly commissioned data center has a press release that reads something like: “Data center achieves unprecedented PUE of 1.15!” That’s a great PUE and something that’s achievable, but is potentially deceiving. Is that a PUE at full load? Is it a snapshot or annualized? What’s the current load (and PUE) of the data center today? What is the data center’s growth plan? How confident is the growth plan? PUE was published by The Green Grid[1] in 2007 as a metric to measure the efficiency of data center facility infrastructure. Since then it has been adopted globally as the go-to metric. PUE can be a snapshot, it can be annualized, it is normally measured, but it can also be predicted. If designers published a new data center’s Design PUE Curve it would drive owners, operators and designers to create more efficient designs over the full load range and my follow-up questions would be unnecessary!
This problem is analogous to the UPS industry 10 years ago. UPS efficiency numbers were published at full load only. Maybe 50% load if you were lucky. Unfortunately, hidden were the crap efficiency numbers at low load levels. Those lightly loaded data centers in a 2N configurations where suffering with UPS efficiencies as low as 60%. But things have gotten better. Most manufacturers publish UPS efficiency curves over the full load spectrum. The EPA adopted UPS energy star ratings that take into account efficiencies across the profile. This transparency has led to great things. Data center designers can better design UPSs into an overall efficient architecture; UPS manufacturers improved their products to better compete, and end users got a better designed and efficient UPS / electrical system.
We need to take this goodness and apply it to the data center as a whole. Imagine if all data centers were specified with a Design PUE Curve? Designs would become more modular. Components with very flat efficiency curves would be leveraged. End users could expect excellent efficiency out of the gate, which relieves some pressure on their operating expenses in the very rare case their growth plan does not match the business plan J. Measured PUE values can be compared to the Design PUE Curve which would lead to continuous improvement in data center design.
This idea is not new! In 2006 Neil Rasmussen published White Paper 143[2] describing data center growth models and benefits in modularity. The paper focused on the capex & risk mitigation benefits, but was updated later to
reflect PUE benefits. This is also articulated in the Data Center Planning Tradeoff Tool[3]. Scott Noteboom of Yahoo advocated for this very thing at an AFCOM Data Center World conference in 2011[4]. It’s now 2016 and I’m writing about it because I am shocked this has not become standard practice over the past 10 years.
Data center owners and operators – It’s up to you!! For your next build or expansion, demand a Design PUE Curve to be published! If it catches on, maybe we can add a new acronym to our vocabulary “DPC” which is actually an acronym within an acronym, making it even better!
Sources:
[1] The Green Grid: https://www.thegreengrid.org/
[2] Schneider Electric White Paper 143: Data Center Projects: Growth Planning.
[3] https://www.apc.com/tool/?tt=8
[4] https://www.datacenterknowledge.com/archives/2011/03/30/pue-time-to-start-grading-on-the-curve/
