For today’s massive Web 2.0 firms, building a private cloud—for either their own or others’ use—can make financial and environmental sense. But what about smaller firms? A study released this week has good news: from an emissions standpoint, a well-designed in-house data center may be greener than going offsite.
Data centers consume 75 billion kilowatt-hours of electricity annually, about 2 percent of the nation’s total energy consumption, which in turn prompted a New York Times expose. However, that article focused primarily on giants such as Facebook and Microsoft. Smaller and medium-sized organizations, which own half of the servers in the United States, might not have the luxury of devoting hundreds of millions of dollars to data center design and build-out—let alone discovering which solutions can produce the greenest results.
In response, The Natural Resources Defense Council, in conjunction with WSP Environment & Energy LLC conducted a study attempting to answer some of these questions. The results were released this week.
The NRDC’s results conﬁrm claims that running a business application in the cloud is generally more energy and carbon efﬁcient than running it on-premise, largely because cloud computing offers greater diversity and can serve more customers at the same time, achieving better economies of scale than small and medium organizations.
“However, this study demonstrates that this is not always the case and exceptions exist,” the report found. “Many factors contribute to the energy- and carbon-efﬁciency of both on-premise server rooms and cloud services. The most efﬁcient server rooms can have signiﬁcantly lower environmental impacts than the least efﬁcient cloud services. Therefore, it is essential for IT managers to consider all the key variables when making application delivery choices based on environmental criteria.”
The study looked at five different data-center scenarios: on-premise equipment that hadn’t been virtualized; the use of co-location; on-premise servers relying on virtualization; private clouds; and public clouds. Optimized public clouds produced the least emissions per user, or under 5 kg CO2e/year; on-premise server farms with no virtualization could generate the most, or more than 45 kg CO2e/year. On-premise facilities also demonstrated the widest range of carbon emissions, ranging as little as 11 kg CO2e/year. That’s well in the range of the worst public clouds, the NRDC found.
Unsurprisingly, the single largest factor to increase the efﬁciency of typical server rooms is higher server utilization, for example through virtualization. But the report also concluded that Power Usage Effectiveness, or PUE, may be given more weight than it deserves: although PUE is still a valuable metric, more emissions may be saved through virtualization or through a greener source of energy than just optimizing cooling and power distribution.
Upgrading server equipment to newer models is another way to reduce overall energy consumption, claimed the NRDC. But applications had to be right-sized: a server whose application consumed only 5 percent of its available computing power might consume something closer to 1 percent after the server underwent upgrades. Again, virtualization would be the answer. The NRDC also found no real differences between the four applications it tested, concluding that other factors held far more sway than just the application used.
Perhaps the most interesting finding for small and medium organizations was that an in-house data center, efficiently cooled, properly managed and virtualized could be greener than an offsite facility. “If the servers are run at low utilization levels and/or powered by dirty electricity, colocation is only marginally better than an on-premise, non-virtualized server room when it comes to carbon emissions,” read the study.
What should server room managers do? The NRDC has two recommendations: first, consider moving your existing computing applications, and deploying new applications, to efficient and low-carbon cloud services. To make the right decision, ask cloud service providers to disclose not just their PUE, but also their average server utilization rate and the carbon emissions factor of their electricity source. Second, the study advised applying energy and carbon efficiency best practices to on-premise servers, including consolidation and virtualization.