The Green Grid recommends the Power Usage Effectiveness (PUE) metric for data centres in order to benchmark and compare energy consumption between data centres and to baseline consumption against which improvements in energy efficiency may be measured.
PUE is defined as:
Total Facility Power
IT Equipment Power
Most data centres operate at PUE = 2 to 3 and some at PUE > 3. PUE = 3 means that for every watt consumed by the computers, 2 watts are being consumed by data centre (or computer room) air-conditioning (CRAC), lighting and power delivery equipment. The goal is thus to get PUE to approach 1. Google, operator of the world’s largest data centres, reports that current state-of-the-art data centres can attain a PUE of 1.2.
We are about to make an attempt to measure PUE for the Stellenbosch University data centre, with help from Dolf Krige and the Measurement & Verification (M&V) Group under Johan Vermeulen at the department of Electronic and Electrical Engineering. As maximum temperatures and IT loads occur in February during the student registration period and start-of-the-academic-year, we need to have the meters installed by then.
The idea is to also create an energy management dashboard for the IT division, which can also serve as a prototype for the campus. Energy efficiency and electricity conservation is critical in this country and if you can’t measure and understand your consumption, then you can’t manage it. The IT division could contribute towards the campus electricity-savings campaign in this manner.
The meters in question are the Web Enabled Energy Meters from Energy Tracking, of which we have purchased three. We may well only be able to approximate a true PUE value, because it may not be possible to place the meters at the necessary points to always distinguish between IT Equipment and non-IT Equipment power and we may not have enough meters either (one meter is measuring the power consumed by the IT office building, leaving two for the PUE measurement). We should be able to supplement permanent readings with judiciously placed, but temporary, loggers from the M&V Group, however.
At worst the measured PUE will be slightly lower than the true value, but it will give us an indication of the potential for improvement and enable us to benchmark.
We realise too that measuring PUE may not be so straightforward for other reasons than only those mentioned above. For instance PUE will change seasonally: during the Cape’s hot summer, which coincides with the start of the academic year, PUE will worsen. Our PUE measurements will be correlated with weather readings from the M&V Group’s weather station close by. It is perhaps here that we could make the best energy-efficiency impacts.
Furthermore, energy losses in system components (e.g. power distribution units (PDUs), UPSs, CRACs) vary with workload. And the data centre system must be viewed holistically as a complex web of interconnected subsystems, an ecosystem if you will. An improvement in one subsystem may lead to an unforseen loss in another elsewhere.
If PUE has a weakness, it is that it only considers energy (or more accurately, power) supplied to the data centre and is silent about the “useful work” that the energy was put to by the computers i.e. it is concerned with energy efficiency, but ignores data centre productivity.
Consequently, the Green Grid proposes a Data Centre Energy Productivity metric (DCeP), where:
DCeP = Useful Work Produced/Total Data Centre Energy Consumed Producing this Work
DCeP will be much more complex to measure and a lot of research is still required to define how it will be measured. The following should be noted:
- DCeP is concerned with energy measurement and not power measurement (as in the case of PUE). Energy consumption by the data centre over a given time period must be measured.
- The key is to quantify “Useful Work” and this is where much research is required. Useful Work must be generally applicable, must address both task runtime and task absolute completion time, a task’s workload can vary with time, and the workload can differ from task to task, amongst others, and above all…
- Useful Work must only be work that is useful to the end customer or the data centre owner.
Ultimately we would want to measure, improve and monitor DCeP.
Gartner. 2008. Green Data Centers: The Six Key Attributes of Data Center Energy Efficiency Metrics (Note: this document is only accessible for SU students and staff: first login here and THEN click on the link above)
The Green Grid. 2007. The Green Grid Data Center Power Efficiency Metrics: PUE and DCiE (pdf)
The Green Grid. 2008. A Framework for Data Center Energy Productivity (pdf)