Changes in software code and improved server utilization could reduce the need for new power plants

While data centers may belong to a relatively new industry, the scale of their impact was quite unknown to me before taking the Energy Technology and Policy class. The numbers are truly enormous. Estimates from the New York Times (some obtained from their internal investigations, others cited from firms in the industry) peg global data center power usage at about 30 GW, amounting to dozens of full-size power plants running full time, with 7 to 10 GW used in the US alone. In total electrical energy, the data center contribution was estimated to be about 2% of US consumption in 2010.  The bigger shock, however, comes when one realizes how little of that energy is actually used. Estimates on utilization were between 6 and 12 percent for most data centers, meaning about 90% of the power is pure waste to idle servers or system loses. In addition, the multiple backups companies require also drain resources and require power use and pollution when tested or used-most data centers have huge banks of backup diesel generators for long outages and huge banks of car batteries or flywheel systems for momentary, millisecond scale fluctuations.1

A sample data center. Image courtesy of the New York Times (Ethan Pines)

One glimmer of hope can be found in software changes from eBay, which operates 52,075 servers that perform 45,914 transactions per kWh and generates $337,000 in revenue per kWh consumed, as of March 2013. An article from Forbes outlined how small changes in software can amount to enormous savings in electricity and cost- eBay cited a small software tweak, slightly reducing the memory used per operation, with reducing their overall consumption by a MW, allowing the removal of 400 servers and saving $2 million. In addition, the company has developed and promoted what they call the “Digital Service Efficiency” dashboard to monitor their internal power use, calculated as the number of operations (transactions and listings) per energy consumed, to find new ways for conservation.2

The eBay ‘Digital Service Efficiency’ dashboard. Image courtesy of Forbes.

Another area with huge potential to curb the ever-growing impact of data centers is changing the habits of those operating them. While technology companies are often viewed as nimble and creative, the mentality for data servers is ultra-conservative-a single downtime event is understandably seen as unacceptable, so most servers are run at full power, around the clock and with multiple backups, and their utilization rate is disregarded. In addition, companies avoid combining servers with other companies or third parties to achieve economies of scale out of fear of losing control. The environmental record is quite checkered for them as well- data centers are regularly cited in hubs like Silicon Valley, northern Illinois, and northern Virginia for violating the Clean Air Act and similar regulations. Amazon, for example, was cited for 24 violations in 3 years (the most major were for not permitting diesel generators or excess pollution from the generators) in northern Virginia and fined over half a million dollars, before settling for $250,000.1 However, there are signs of hope too-eBay made these software improvements and consolidated to 2 server types, while and Google and Facebook both designed their own, improved servers for themselves.

For me, it seems clear that there may be much to gain in power conservation from both software and utilization improvements. On the software side, eBay’s move seems to reflect a return to the efficiency mentality more common in the 1980s and 1990s, when slow processors and tiny RAM storage required creative and efficient coding to conserve every possible computation. In addition, finding new ways to improve server utilization seems absolutely necessary to support the industry’s growth. This may mean developing new technology like instant power up from sleep mode and larger solid state drives, reducing redundancies, or investing in economies of scale by joining forces in ever-larger individual data centers. All told, the idea of utilizing software upgrades and habit changes to avoid capital investments in the electrical grid is certainly an intriguing option.


1)      The New York Times, James Glanz,  The Cloud Factories-Power, Pollution, and the Internet, Sept. 22, 2012.

2)      Forbes, Michael Kanellos, EBay’s MPG for Data Centers, March 21, 2013.


1 Comment

Filed under Uncategorized

One response to “Changes in software code and improved server utilization could reduce the need for new power plants

  1. This is an interesting point and leads to the question about the efficiency of large scale technology companies. e-Bay is one example, but there are numerous others (Amazon, Google, Yahoo!, etc.) that require the use of many servers. Are these efficiently run? And what about other high-technology companies, such as those in the biomedical field or the oil and gas industry. Surely these also have small inefficiencies that can be eliminated to reduce energy consumption.

Leave a Reply

Please log in using one of these methods to post your comment: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s