Peter Hopton, green IT entrepreneur and Iceotope founder, says the popular data centre efficiency metric PUE is being abused by marketeers, rendering hollow figures rather than useful efficiency measurements.
Everyone in the data centre industry and a surprising number of people outside of the industry can now tell you that PUE is a measure of energy efficiency in a data centre and that lower is better.
PUE or Power Usage Effectiveness is an instantaneous measurement of the ratio between total power consumption and IT power consumption in a data centre. As a result, PUE can and does vary over the year.
Justifying a low PUE rating
You also might find low PUEs being claimed by organisations that look to ignore or subtract some loads from their 'total power consumption’ such as measuring power consumption on the low voltage side of the transformer – saving them 0.05 off their PUE. Other tricks include turning the lights off or switching the UPS into bypass or line interactive mode when taking the measurements, but then flicking it back into double conversion shortly after as they don’t trust it will kick in fast enough in a power event.
• Some organisations are reducing their data centre cooling energy, but specifying servers with larger fans – moving the cooling energy into the IT load. It is estimated that as much as 20 per cent of IT load is fans moving air through the device to cool it
• Facebook uses lighting run over POE (power over Ethernet), which means the lighting and energy transmission losses are included in the IT load. A 25W POE device requires about 40W of power at the POE switch’s input, the remaining 15W is lost as heat over the cable or in the switch PSU
With the Green Grid having no plans to certify or involve itself in disputes over the PUE metric manipulations are only going to increase, so the next time you see a PUE claim that looks too good to be true, it probably is.
PUE or Power Usage Effectiveness is an instantaneous measurement of the ratio between total power consumption and IT power consumption in a data centre. As a result, PUE can and does vary over the year.
So strictly speaking if you report your PUE publicly, you should take the average over the year, no? You might, however, find that reported PUE’s represent the system’s best performance or worse – its theoretical best performance. This 'design PUE' is common in new unpopulated data centres which will have a very high PUE due to the fixed load of the cooling outweighing the insignificant IT load, a new data centre in these conditions could have a PUE of 7, which will settle down to 1.6 after it becomes populated.
Justifying a low PUE rating
Prefixing PUE doesn’t just stop at design modelling, you may find people talking about pPUE, mPUE or some other prefix that isn’t defined in the standard. This is because they are attempting to define a particular set of conditions outside of the standard of the metric in order to justify a low PUE claim. Some experts in the industry have suggested that in the case of mPUE, the m simply stands for marketing!
You also might find low PUEs being claimed by organisations that look to ignore or subtract some loads from their 'total power consumption’ such as measuring power consumption on the low voltage side of the transformer – saving them 0.05 off their PUE. Other tricks include turning the lights off or switching the UPS into bypass or line interactive mode when taking the measurements, but then flicking it back into double conversion shortly after as they don’t trust it will kick in fast enough in a power event.
And then there are designs that don’t break the standard that is PUE, but find interesting engineering work-arounds in order to reduce it:
• Some organisations are reducing their data centre cooling energy, but specifying servers with larger fans – moving the cooling energy into the IT load. It is estimated that as much as 20 per cent of IT load is fans moving air through the device to cool it
• Facebook uses lighting run over POE (power over Ethernet), which means the lighting and energy transmission losses are included in the IT load. A 25W POE device requires about 40W of power at the POE switch’s input, the remaining 15W is lost as heat over the cable or in the switch PSU
With the Green Grid having no plans to certify or involve itself in disputes over the PUE metric manipulations are only going to increase, so the next time you see a PUE claim that looks too good to be true, it probably is.
Like this content? Please subscribe to our free weekly e-newsletter at the top of the page for more content like this.
Related content:
23rd August 2012
No comments:
Post a Comment