22 Jul 2009

22 Jul 2009

LinkedIn Discussion on Power Usage Effectiveness (PUE)


Category: Uncategorized

Last week I posted the following discussion question in our Computer Room Design networking group at LinkedIn.com. I’m really impressed with the response from group members, so I’d like to share their thoughts with you here:

How can the industry address problems with the reporting of Power Usage Effectiveness (PUE) without undermining the usefulness of the metric?

In a recent post in Data Center Knowledge, Rich Miller points out that the value of Power Usage Effectiveness (PUE) as the leading ‘green data center’ metric “has become fuzzy due to a disconnect between companies’ desire to market their energy efficiency and the industry’s historic caution about disclosure.” [Source: http://www.datacenterknowledge.com/archives/2009/07/13/pue-and-marketing-mischief/]

What are your thoughts on redefining PUE? Are additional refinements the answer? Or does increasing the complexity of PUE undermine the usefulness of the metric?


• Gordon Lane, Facilities Coordinator at Petro Canada, explained:

I don’t see a real value in PUE.

If you leave unused servers powered on you can keep your PUE low.

Assume you have a PUE of 2
2MW total power consumption gives you 1 MW for servers.
If you can reduce your server consumption to 0.75MW by turning off comatose servers total consumption reduces to 1.75MW and gives you a PUE of 2.33

I know there would be some reduction in a/c power usage due to less heat output from the turned off servers but if you are using legacy a/c units with no VFD style control then you will not get a corresponding electrical consumption reduction.

• Scot Heath, Data Center Specialist, weighed in with:

PUE is difficult to measure in mixed facilities, is muddied by configurations such as the Google every-server-has-a-battery and varies widely with Tier level. A universal measurement that combines both IT capability (total Specmarks for example) and availability with respect to energy consumption would be most useful. PUE does have the advantage of being quite easily understood and for controlled comparisons (like tier level, etc.) is very useful.

• Dave Cole, Manager of Data Center Maintenance Management and Education Services at PTS, responded:

Gordon and Scot bring up very good points. I have mixed feelings about PUE. The concept is easily understood – we want to maximize the power that is actually used for IT work. The interpretation of the value is easy to understand – lower is better (or higher is better in the case of DCiE). The problem I see is that it’s almost been made too simplistic. You still have to know your data center and the impact of the decisions you make in regards to design and operation. You can actually raise your PUE by virtualizing or by turning off ghost servers as Gordon pointed out. What needs to be understood is that when you lower the demand side, you should also be making corresponding changes to the supply side. At the end of the day, PUE can be valuable as long as you are also looking at what impacts the value. You need to be able to answer the question of WHY your PUE is changing.

What are your thoughts on the value of Power Usage Effectiveness (PUE) as a metric? Please share your experience by posting a comment here, or by continuing the discussion in the Computer Room Design Group on LinkedIn.


Comments (3)

  1. Andrew Graham on 27th Jul 2009

    Even the Green Grid must recognize the limitations & flaws in the effectiveness of PUE as a measurement for data center efficiency as they have released 3 new levels of PUE, Basic, Intermediate & Advanced as reported by Rich Miller on Data Center Knowledge.


    I would like to get opinions on whether these new levels will eliminate the flaws & limitations discussed above? Since the new Advanced Level is measuring at the server level continuously that should actually help us shut down comatose servers to improve our PUE.

    Also who is interested in using the new Advanced Level & how do you plan to get there? There are some effective rack PDU's & software to measure at the server level. There are also tools to measure power usage from IPMI, but that of course will require all of your servers & IT infrastructure to have IPMI available.

  2. Ken Oestreich on 27th Jul 2009

    Nice discussion. I would simplify the discussion in the following manner:

    PUE is a great metric for facilities managers b/c it helps drive-out inefficiencies in data center infrastructure. However (as Gordon points out) it doesn't include any metric on how efficient the computing resources are themselves.

    So, we need a -second- metric that can be used alongside PUE that targets IT itself. Even something as simplistic as average server utilization would be a helpful metric to use alongside PUE.

    In that way, you get a fuller picture of infrastructure efficiency *and* compute resource efficiency.

  3. Tim Dodson on 31st Jul 2009

    As a technician, I'm always aware that testing characteristics define the result, given that, the 3 levels of PUE should better assist defining how we view utilizing our data center power structure. Power usage will always fluctuate according to server/MF/san utilization, powering down any equipment can be effective, but still needs to meet client service level agreements.

Comments are closed.