Earlier this fall, Google made an announcement that in many ways foretells the future of data center efficiency metrics. The search giant not only disclosed its total power consumption and carbon emissions (mostly attributable to its data centers), but also released estimates of its per-user and per-search energy consumption. With this information -- and given that a billion Google searches occur per day -- it was possible to calculate that searches account for 12.5 million watts of the company's 260 million watt total.
The idea of quantifying the kilowatts of energy required to perform a useful unit of work is now considered the holy grail of data center metrics. The most widely used metric today -- The Green Grid's Power Usage Effectiveness (PUE) measure -- compares the amount of data center energy consumed by IT equipment to the facility's total usage; but no metric yet reveals energy consumed per unit of work, mostly because it is difficult to define a standard "unit of work."
"For cars, it's miles per gallon, but for hardware, is it a MIP or bits transmitted per second? The industry hasn't zeroed in on what 'work' is," says Steve Press, executive director of data center facilities at Kaiser Permanente. "PUE doesn't tell you what work the data center is doing -- it could be blocking virus traffic or moving data for the National Weather [Service]. Once we settle on [that], it will have a big effect on overall competitiveness."
Power Usage Effectiveness (PUE)
PUE is by far the most widely used data center efficiency metric. Developed by The Green Grid, it illustrates how much energy in the data center is allocated to IT equipment -- servers, networks and storage -- and how much is used for cooling, lighting and power equipment. Its popularity is attributable in part to its simplicity, both in concept and in application.
PUE is calculated by taking total facility energy (cooling, lighting, power and IT) and dividing it by IT equipment energy. A "perfect" PUE score, then, would be 1. According to The Green Grid benchmarks, a PUE of 2.2 indicates "needs improvement," while 1 to 1.4 is best in class. Some organizations prefer to use the Data Center Infrastructure Efficiency (DCiE), metric, which is the inverse of the PUE ratio.
"Despite its very well-publicized limitations and frequent misuse, PUE is a good measure when it is clearly understood by those using and interpreting it, and when it is measured in a consistent way," Mingay concludes.
Carbon Usage Effectiveness (CUE) andWater Usage Effectiveness (WUE)
CUE, recently released by The Green Grid, addresses data-center-specific carbon emissions. It is calculated by measuring total CO2 data center emissions and dividing by IT equipment energy. WUE, also a relatively new metric, measures the amount of water used for IT and is similarly calculated by measuring total water usage in the facility and dividing by IT equipment energy.
For example, if an email server is processing emails 75% of the time, then its ScE would be 75%, Mingay says. "It looks at not just 'Is your system doing something?' but 'Is it doing something useful?' " Stanley adds. The idea is to identify underutilized servers and apply virtualization, consolidation or decomissioning strategies to increase efficiency.
The Green Grid emphasizes that these metrics are not productivity metrics -- they don't measure how much work is done, but rather what proportion of work is useful. That distinction is important, Mingay says, since a server with a slow processor might spend an hour processing 100 transactions, while a server with a fast processor might complete 100,000 transactions in that time. However, both would get an identical ScE of 100% if 100% of their time during the hour was devoted to doing that work.
Data Center Energy Productivity (DCeP)
Something to keep in mind is that there isn't a big drop-off in power consumption for servers running at low capacity, Schirmacher says, so even if you purchase servers with good ratings, you also need to ensure you're achieving high utilization, which is measured by ScE.
Gartner Power to Performance Effectiveness (PPE)
Gartner developed PPE to analyze the effective use of power by existing IT equipment, relative to the performance of that equipment. With PPE, companies measure the average performance, energy consumption and utilization of their equipment and then compare that with optimal targets they have previously defined. PPE is represented, then, as actual power performance, divided by optimal power performance. "Pushing IT resources toward higher effective performance per kilowatt can have a twofold effect of improving energy consumption -- putting energy to work -- and extending the life of existing assets through increased throughput," Mingay says.