Study Quantifies Cloud’s Energy-Savings Potential
There is mounting evidence that the cloud is a green cloud. A new study from Lawrence Berkeley National Laboratory and Northwestern illustrates the Internet's potential for saving energy.
The Google-funded study measured the effect of moving three common business applications – email, customer relationship management software, or CRM, and bundled productivity software – to the (public) cloud. The results were dramatic. If all 86 million US workers migrated to the cloud, the energy needed to run those applications would decrease by about 87 percent. That is an amount equal to 23 billion kilowatt-hours of electricity annually, enough to power the city of Los Angeles for a year.
Green Computing Report spoke with the principal investigator for the project, Berkeley Lab's Lavanya Ramakrishnan, and Northwestern's Eric Masanet, lead author of the report. They explained that although the research project received financial backing from Google, it employed generic, vendor-neutral data, drawn from a mix of sources including US Census data.
The researchers detailed their findings in a thorough 38-page report, which contends that the energy implications of moving to the cloud are affected by multiple dynamics.
Using a systems-based approach, the team developed an open-access model, called the Cloud Energy and Emissions Research Model, or CLEER. CLEER takes into account the energy footprint of datacenters, data transmission systems, client IT devices, commercial and residential buildings, as well as manufacturing, transportation, and waste management systems. It also includes key interrelationships between these end use systems.
As elucidated in the report, "the CLEER Model quantifies the net changes in regional energy use between present day and cloud-based systems – accounting for changes in both direct energy use and embodied energy – and calculates the resulting net changes in direct and embodied GHG emissions."
For the researchers' large-scale case study, business applications were selected for their near-ubiquity. Email, productivity software (think Windows or Google Apps), and CRM software are currently used by tens of millions of US workers.
The upper bound for the present-day energy footprints of the three business applications was estimated at 373 petajoules per year. (One petajoule is equal to 10^15 joules.) Datacenter operations made up the biggest percentage (86 percentage) of this energy footprint, followed by the operational energy use of client IT devices. According to the model, running the same application sets in the cloud would only consume about 47 petajoules each year, a savings of 326 petajoules.
The researchers concluded that the energy savings was mainly the result of consolidation. Their report refers to the "vast differences between the energy efficiencies of local and cloud datacenters." The shift to shared resource pools – aka public clouds – has led to the creation of larger and larger datacenters. According to an IDC study, service providers will soon account for more than a quarter of all large datacenter capacity in the US. There are far fewer datacenters overall, but the ones that remain are larger and far more efficient.
"Combined, high server utilizations and low PUEs have made cloud computing the new standard for best practice datacenter energy efficiency," the report's authors write.
The main research tool used in the case study, the CLEER model, is also freely available to the public. Now anyone can capture the net energy implications of migrating applications to the cloud. The open-access model allows users to input scenarios and build a custom case study comparing their present setup with a cloud-based implementation.
Northwestern's Eric Masanet talked about the project at Google's How Green Is the Internet? event last week. He was the first presenter in a rapid-fire research session called "Business software as a service: how much energy can it save?"
Masanet remarked that a lot of these models come with caveats - the savings cannot be predicted with absolute certainty because there is a chronic lack of data concerning the server and datacenter space, and the data they used was drawn from various sources with different parameters. As with all models, assumptions are made. That's why transparency is so important; all of input values and all of the assumptions in this model are transparent. Masanet welcomes the research community to analyze the public use model, to change the assumptions they don't like and improve the model over time.
"But the primary conclusion does not change," he adds. "Even in the face of these uncertainties, a shift to the cloud – moving away from inefficient datacenters toward highly efficient and far fewer datacenters – is likely to save a lot of energy."
While the CLEER model was designed for the energy analysis research community, the study authors note that it will also serve as a resource for business and policy experts seeking to better understand the environmental implications of transitioning to cloud-based services.