Companies that own numerous data center operators across the globe could be able to save millions of dollars a year in electricity costs if they dynamically shifted computing power across their data centers to when and where energy prices are the cheapest. At least that’s according to a study out this week from the Massachusetts Institute of Technology and Carnegie Mellon (hat tip Ars Technica).
In other words, companies that have lots of data centers can take advantage of cheap bandwidth, smart software and fluctuating hourly energy prices to shift computing power to a data center in a location where it’s an off-peak time of the day and energy prices are low. Commonly that’s in the middle of the night, which is why industry-watchers like Rich Miller, editor of Data Center Knowledge, call the process “following-the-moon.”
The MIT and Carnegie researchers looked at hourly electricity prices for 29 locations in the U.S. over a period of 39 months, as well as data collected off of Akamai’s (s AKAM) data network, to create a simulation for how much companies could save by shifting computing loads. The group found that companies could save over 2 percent in energy costs without a substantial rise in bandwidth costs or a drop in performance for computing services.
For data centers that are more “energy proportional” — using energy efficiently across a range of activity levels, from idle to peak load, as I explained on GigaOM Pro (subscription required) — and don’t have any constraint on bandwidth use and speed, the savings could be as high as 13-30 percent. Most data centers, however, are not very energy proportional, and companies that depend on super-fast on-demand bandwidth, would likely tend to prioritize connection speed above cheap energy more than another company selling services that depend less on scalable available bandwidth.
Still, while energy savings of 2 percent might not be a lot for a small company with a modest energy bill, for a company like Google (which the researchers predict spends $38 million per year on data center energy costs) or Microsoft (s MSFT) (estimated to spend $36 million per year) the savings could be substantial. Close to $1 million each for Google and Microsoft based on these estimates.
The study is interesting because while some companies with massive distributed data centers are starting to employ these tactics (data center software maker Cassatt, for example, sells a product that dynamically shifts loads to find the cheapest energy prices), this is still a relatively new concept. It’s particularly interesting for companies that offer cloud computing services, selling scalable on-demand computing as a service, since they could use their massive networks to create significant savings and pass that onto their customers.
Given many cloud computing providers are already shifting computing loads to different locations to provide fast delivery and on-demand bandwidth, the researchers suggest that adding in an energy price cost policy wouldn’t be that difficult. And as longtime IT energy researcher Jonathan Koomey found in one of three reports released this week, cloud computing companies are already leading the charge in being smarter about energy use.
But there are drawbacks to the concept. Shifting computing loads to find cheaper energy prices actually could cause the amount of energy required for computing to rise slightly, because a company is using more bandwidth. In addition cheaper energy prices are currently delivered by some of the dirtiest fossil fuel options — namely coal — so I would predict it could boost the company’s use of the dirtiest/cheapest form of electricity.
An alternative, which Stacey on GigaOM has written about, is helping data centers shift computing loads to tap into renewable power. But unfortunately for the time being, until clean power drops in price, that’s not going to save you a whole lot of money.
The one point of defense that I could think of from a climate change perspective for load-shifting to find cheaper power is that perhaps this type of load-shifting could help utilities manage the grid better. If utilities better manage the grid then they don’t have to build as many dirty peaking power plants. I would urge the team to do more research in this area.
Image courtesy of Flickr Creative Commons.