Cleantech Market Intelligence
A FVER Cure for Data Center Inefficiency
An article in The New York Times on data center energy consumption has brought the debate over the hidden environmental costs of the digital society to a broader audience. We have examined several of these issues in previous blogs and in our green data center report.
I think the article underplays the efforts the IT industry is making to encourage best practices on energy efficiency, notably through the work of organizations like The Green Grid. But it does highlight one particularly intractable problem: the unproductive consumption of energy by IT systems. It’s a mark of shame for the industry that typically only 6% to 12% of the energy consumed by IT systems is used to do useful computational work. This continues to be a problem despite campaigns to get data center managers to turn off idle servers and the availability of tools that can help monitor and manage the power use of IT.
One of the barriers to improvement has been the lack of an accepted metric that enables data centers to be compared in terms of their productive use of energy. PUE (Power Usage Effectiveness) is now well established as a metric for data center energy efficiency, but it doesn’t capture data center productivity. In fact, running your IT kit at full power even if it is not doing effective work will improve your PUE score (as IT would be a greater percentage of your energy consumption).
There are several current initiatives to produce a new metric that will account for the productive element of IT energy consumption, giving us a means of assessing which data centers are truly energy efficient. The Green Grid has produced an initial model and work is also being done in the EU and in Japan, but it’s proving difficult to produce a metric that is both insightful and simple, which has been key to the uptake of PUE. Now, though, a new metric developed by the BCS Data Centre Specialist Group may have cracked the problem.
Fixed vs. Productive Power
Called Data Centre Fixed to Variable Energy Ratio, or DC-FVER, it can be used to assess the amount of energy consumed for productive work compared to fixed energy consumption (that is, the energy consumed even when the servers are delivering little or no business value). FVER allows each data center to define a useful work metric that is relevant to a company’s business, so it’s tailorable to different types of data center. For example, a news website might measure productivity in terms of the number of webpages served per hour, while an online retailer may look at the number of transactions per second. FVER is defined by the formula:
FVER = 1 + (Fixed Energy (kWh)) / Variable Energy (kWh))
As with PUE, the ideal FVER rating is 1, which would mean all energy is being consumed for productive work. In reality, most data centers are expected to have scores closer to 10 on first assessment.
A simple example given by Victor Smith, Vice Chairman of the BCS, at the Green Data Centre Conference explains how it might work. After choosing a useful work metric for a specific data center, the output and the energy consumption needs to be measured, say, every hour for a week. At the time of minimal work, it’s possible to get a measure of the data center’s fixed energy consumption. So if in the most productive hour 1.2 megawatt-hours (MWh) are consumed and in the least productive time 1 MWh, then variable consumption is 200 kWh, (the energy consumed for productive work). In this case:
FVER = 1 + (1000 kWh / 200 kWh) = 6
FVER will also allow direct comparison between data centers based on their productivity relative to energy consumption. FVER is only one among a number of approaches to measuring data center productivity, and more complex metrics will also have their place, but it could help accurately assess the efficiency of data centers.