Which Geographies Use the Most Juice for Servers?
January 14, 2008 Timothy Prickett Morgan
Back in February 2007, Jonathan Koomey, a staff scientist at the Lawrence Berkeley National Laboratory and a professor at Stanford University, created a model of server installed base and electricity usage for these servers based on data from IDC to try to get a handle on how much power servers were consuming and how quickly that consumption was growing. The data in Koomey’s original model was based on IDC stats for the United States and the world as a whole, and he has now updated it with finer-grained data from IDC to show power use by geographical regions.
The original study back in February estimated that in 2000, the United States had some 5.6 million servers installed and that they consumed 23.3 billion kilowatt-hours of juice that year, including the electricity for the servers themselves and their share of power transformation as wires come into the data center and energy consumed to cool those servers. Worldwide, the 14.1 million servers installed consumed 58.4 billion kilowatt-hours of juice in 2000, giving the U.S. a 39.9 percent share of the juice consumed running and cooling servers. Stepping up to 2005, Koomey estimated that the 10.3 million servers in the U.S.–up 83.9 percent in six years by box count–consumed at total of 45.1 billion kilowatt-hours of electricity–up 93.6 percent over the six years. The growth in electricity usage grew faster than the installed base in this time, which is impressive given the level of server consolidation over those six years. On a worldwide basis, Koomey estimated that the 27.3 million servers installed at the end of 2005 (up 93.6 percent compared to 2000) had consumed 122.9 billion kilowatt-hours of electricity in the year (up 110.4 percent). Obviously, the server base and their power consumption on a worldwide basis was growing faster over those six years, which begged the further question of where this growth was occurring.
So, Koomey tried to answer it with some more estimates, based again on IDC shipment data that breaks the world down into the United States, Western Europe (the usual suspects), Japan, Asia/Pacific (excluding Japan), and the rest of the world (which includes Eastern Europe, Latin America, the Middle East, and Africa). It is unclear from the IDC Worldwide Server Tracker service where Canada went, but importantly, India, China, Indonesia, Singapore, Australia, New Zealand, and a number of fast-growing countries in the AP region are lumped together in this data.
The upshot of this revisiting of the power usage of servers, which is paid for by chip maker Advanced Micro Devices and which is available as a report on the AMD site, is that the United States is the power hog that you would expect it to be, and that is based mainly on the large installed base of servers that U.S. data centers, data closets, and spaces under desktops have humming away inside of them. In 2005, Western Europe had 7.4 million servers that consumed 33.3 billion kilowatt-hours of electricity, which represented a 93.9 percent growth in the server installed base and a 120.5 percent growth in power consumption in those countries over the six years from 2000 to 2005. Japan had 2.6 million servers at the end of 2005, up 82.1 percent, and power consumption compared to 2000 rose by 92.5 percent to 12.9 billion kilowatt-hours. The Asia/Pacific region had 3.7 million servers in 2005, up 138 percent compared to 2000, and their aggregate power consumption rose by 175.9 percent to 16 billion kilowatt-hours per year by 2005. The rest of the world, which includes Russia, Brazil, and a number of vibrant economies, accounted for 3.4 million servers in its installed base in 2005 (up 92.5 percent from the base in 2000 six years earlier), and power consumption for these machines in 2005 came to 15.5 billion kilowatt-hours, up 103.9 percent.
The amazing thing is that China, India, Russia, Brazil, and other fast-growing economies are not seeing their energy consumption in the data center rise faster than this latest report from Koomey suggests.
But in many cases, the growth in the economy is in an area that is not heavy in computing–such as manufacturing or oil production–compared to the established financial services, online, and distribution companies in the Western economies, which certainly are heavy users of computers. The other thing to remember is that in fast-growing economies, everything is a scarce resource, and electricity is no exception–especially in China and India. According to some IT vendors, the limit of available power and readily available human labor is what is keeping companies from computerizing more than they do. But that could change rather quickly. The Chinese and Indian economies were forced by the circumstances of their relative poverty to skip land-based telephone lines and move to straight to cell phones, and the fact that these countries are, relatively speaking, undercomputerized compared to the economic might they now wield also suggests that they might be able to skip over the inefficient, expensive computing that many Western companies are still saddled with and move right to energy-efficient server and storage designs from the get-go. It will be interesting to see what happens as more and more companies computerize in developing countries, in large measure because modern computer designs will allow them to do a lot of work with a lot less power.
Koomey did a little prognosticating in his report, and estimates that by 2010, the share of global electricity consumed by servers in the United States could drop from just under 40 percent to around 34 percent, while the Asia/Pacific region’s share of juice consumption will rise from 10 percent to 16 percent. And it looks like it will be something akin to a direct transfer, since Koomey is guessing that the share of the global electricity consumption by servers in Western Europe, Japan, and the rest of the world will remain about the same. Koomey expects that the absolute amount of electricity consumed in the Asia/Pacific region will more than double again between 2005 and 2010, compared to a 76 percent growth rate for the United States in the same term. Those estimates, of course, assume that server platform shipments grow as in the past and power consumption per server remains roughly the same per box in a general class–volume, midrange, and high-end.
Neither of these may turn out to be a valid assumption, particularly if the key global economies turn south. Then, all bets are off and a major transformation–perhaps to compiled programs running on cheap, skinny machines, perhaps to utility computing–will be underway. We’ll just have to wait and see.
RELATED RESOURCES AND STORIES
Estimating Regional Power Consumption by Servers: A Technical Note, Jonathan Koomey