WHAT IS PUE?!
Power Usage Effectiveness (PUE) is a hot topic in data centers, especially with different countries implementing energy regulations regarding PUE.
It is a ratio of the TOTAL POWER CONSUMED BY THE DATA CENTER divided by the IT POWER CONSUMED IN THE DATA CENTER. This results in a dimensionless number greater than 1, a less than 1 PUE is simply not possible without massaging the numbers with renewable energy generation. The closer the PUE to 1.00, the more effective a data center is using power for IT load.
It seems sometimes simplified, but it can be very complex. The common thought is the best (most effective) PUEs are closer to 1.00, and it is partially right. Firms design data centers to keep PUE as low as possible, and advertise greatly on this low PUE if they can attain it. A consideration with this is also lowering the total power used by data centers, but in this age of Artificial Intelligence and Machine Learning (AI & ML) this may be less of a thought as rack densities are ever increasing.
There are multiple parts of PUE, and they can change based on the measuring points:
PUE 0 :
The instantaneous peak load of the total data center power divided by the UPS output instantaneous peak load consumed
PUE 1 :
Total annualized kwh of the DC divided by the total annualized kwh consumption of the UPS at the output of the inverters
PUE 2 :
Total annualized kwh of the DC divided by the total annualized kwh consumption of the Power Distribution Units (PDUs) at the output of the PDUs.
PUE 3 :
Total annualized kwh of the DC divided by the total annualized kwh consumption of the IT equipment on the input to the racks.
Where can PUE be improved the most?
Let’s take an example of a small data center:
-Total Consumption : 800kW
-Total IT : 500kW
-Mechnical cooling : 240kW
-Power distribution and losses : 35kW
-Lighting and office loads : 15kW
-Fresh air ventilation : 5kW
-Security Systems : 3kW
-Communications equipment 2kW
This is a PUE of 1.6. Not bad, and we can see the biggest load after IT equipment is the cooling. Improvements in cooling plant and equipment efficiencies can really improve PUE, since it is lowering the largest contributor to the total DC consumption. Using different cooling methods (adiabatic cooling, free cooling, chilled water cooling), lowering fan speeds on CRAHs to match airflow demands, raising the chilled water temperatures, improving containments systems in aisles and underfloor systems, these can all greatly lower the mechanical loading of the building to cool the IT loads. In an example for a raised floor DC, an operator and their team simply adjusted their cold aisle floor grate dampers to match the rack it served and installed bushings on previously open floor penetrations to electrical equipment, and they lowered PUE from 1.26 to around 1.22, a massive improvement for this large DC! Simple changes and choices in design are critical for setting and improving PUE. Even lighting loads can make the difference.
The future of PUE is also changing. With increases in liquid cooling methods, designers are pushing the limits of PUE improvements. There is also an interesting paper about something called TUE (Total Usage Effectiveness), which even more granularly, separates the infrastructure PUE and servers fans from the actual compute equipment of a server. Generally it is two parts:
ITUE : TOTAL POWER CONSUMED BY THE DATA CENTER but only considering IT(computer) all divided by the TOTAL IT used
PUE : TOTAL POWER CONSUMED BY THE DATA CENTER divided by the IT POWER CONSUMED IN THE DATA CENTER.
TUE = ITUE x PUE
Much more on this on the web: https://datacenters.lbl.gov/sites/default/files/isc13_tuepaper.pdf
All my articles are handwritten/typed by me. By reading and sharing these, you support this work and you support real human authorship.