TODAY’S STUDY: HOW DATA CENTERS CAN SAVE THE U.S. $3.8 BIL A YEAR
Data Center Efficiency Assessment; Scaling Up Energy Efficiency Across the Data Center Industry: Evaluating Key Drivers and Barriers
August 2014 (Natural Resources Defense Council)
Data centers are the backbone of the modern economy, from the server rooms that power small- to medium-sized organizations, to the enterprise data centers that support American corporations, to the server farms that run cloud computing services hosted by Amazon, Facebook, Google, and others. However, the explosion of digital content, big data, e-commerce, and Internet traffic is also making data centers one of the fastest-growing users of electricity in developed countries, and one of the key drivers in the construction of new power plants in the United States.
While most media and public attention focuses on the largest data centers that power so-called cloud computing operations—companies that provide web-based and other Internet services to consumers and businesses—these hyper-scale cloud computing data centers represent only a small fraction of data center energy consumption in the United States.
As NRDC initially found in its groundbreaking 2012 analysis, Is Cloud Computing Always Greener? Finding the Most Energy and Carbon Efficient Information Technology Solutions for Small- and Medium-Sized Organizations, smaller server rooms and closets are responsible for about half of all U.S. server electricity consumption—but 50 percent of that is wasted due to lack of awareness and incentives to make them more efficient. There remains a critical need for action, including developing utility incentive programs to reduce waste in the massive amounts of electricity used by data centers small and large.
In 2013, U.S. data centers consumed an estimated 91 billion kilowatt-hours of electricity. This is the equivalent annual output of 34 large (500-megawatt) coal-fired power plants, enough electricity to power all the households in New York City twice over. Data center electricity consumption is projected to increase to roughly 140 billion kilowatt-hours annually by 2020, the equivalent annual output of 50 power plants, costing American businesses $13 billion per year in electricity bills and causing the emission of nearly 150 million metric tons of carbon pollution annually.1
If just half of the technical savings potential for data center efficiency that we identify in this report were realized (to take into account the market barriers discussed in this report), electricity consumption in U.S. data centers could be cut by as much as 40 percent. In 2014, this represents a savings of 39 billion kilowatt-hours annually, equivalent to the annual electricity consumption of all the households in the state of Michigan. Such improvement would save U.S. businesses $3.8 billion a year.
There has been significant progress in data center efficiency over the past decade, with shining examples of ultra-efficient server farms run by the likes of Google and Facebook. But how efficient, really, is the typical data center in the United States? What are the key opportunities for further efficiency gains, and what are the main barriers to capturing these opportunities? Further, while projections of overall industry growth and its energy and carbon impacts remain elusive, even less is known about the makeup of the sector by data center type and the impact this may have on future carbon emissions. These are the questions that NRDC and Anthesis set out to answer in this extensive survey by interviewing more than 30 industry stakeholders and experts and reviewing the latest industry literature.
Much of the progress in data center efficiency over the past five years has occurred in the area of facility and equipment efficiency. However, little progress has been achieved in server operation efficiency, particularly in terms of server utilization. In addition, progress remains uneven across the different segments of the data center market. This study therefore focuses on assessing the current situation, the opportunities and the barriers related to server utilization efficiency, and the ways in which these opportunities and barriers vary across different segments of the data center market, particularly with regard to the multi-tenant data center business model.
This study found that while the largest public-facing companies providing cloud computing services typically run their data centers very efficiently, progress on energy efficiency has slowed in other types of data centers. Most significantly, servers are being used very inefficiently, consuming power 24/7 while doing little work most of the time. This is due to a variety of factors, such as these:
• Peak provisioning: Data center operators install enough equipment to handle peak annual load, and then some, but do not power down unused equipment during the majority of the time when it is not needed.
• Low deployment of virtualization technology (which allows the consolidation of workloads onto fewer servers) across the entire server fleet.
• “Comatose” servers: A large number of servers that are no longer being used still gulp energy 24/7 because no one is decommissioning them or is even aware that they’re no longer used.
Multi-tenant data centers are shared data centers where customers lease space and power to run their computing equipment rather than managing their own data center.
This creates a potential split incentive situation, where the decisions made by customers on the efficiency of their information technology (IT) equipment have no direct impact on their bills, removing any financial motivation to improve energy efficiency.
Split incentives, both internally within organizations and between multi-tenant data center providers and customers, remain one of the biggest reasons why efficiency best practices are not being implemented at scale across the data center industry. Only 20 percent of organizations’ IT departments pay the data center power bill, a statistic that has not changed in five years. While this issue has been largely solved among hyper-scale cloud service providers, contractual relationships between colocation providers and their customers compound the split incentive challenge, with the data center owner paying the power bill, the tenants buying power blocks, and their IT purchasers separately specifying equipment.
The technical solutions, such as virtualization, server power management, and single responsibility for IT and facility costs, are well known but are not applied systematically. This report recommends system-level policy actions to create the conditions for best-practice efficiency behaviors across the data center market. One of these systemic levers is the adoption of a simple CPU utilization metric that would provide internal and external visibility to the IT efficiency of data centers, creating management and market incentives for operators to optimize the utilization of their IT assets. Others include public disclosure of data center energy and carbon performance, and the development of templates for green colocation and other multi-tenant data center contracts.