Google has tested a demand-response system, which will reduce data center electricity consumption during times of stress on local electricity grids, by shifting non-urgent computing tasks to other times and locations.
The company says this function is an extension of an existing program to shift computing work to where the energy is least carbon intensive, which was first tested in 2020.
The system has now been used to alleviate demands on local grids in Taiwan, Oregon, and across Europe, and is planned to be rolled out more widely. The company assures customers that it will not impact Google services.
"Since 2020, we’ve used our carbon-intelligent computing platform to shift compute tasks and their associated energy consumption to the times and places where carbon-free energy is available on the grid," says a blog post from Varun Mehra, technical program manager grid services, and Raiden Hasegawa, senior data scientist, energy analytics.
"Now, we can use this task-shifting capability for demand response — temporarily reducing power consumption at our data centers to provide valuable flexibility when it is needed, to help local grids continue operating reliably and efficiently."
According to the blog, utilities are often forced to add new fossil fuel capacity to grids that are frequently overloaded. Demand response has been practiced for some time, but only rarely, and usually as an emergency measure. Data centers are also often reluctant to change their operations at the behest of the utility.
In recent weeks, the US Department of Energy has published research suggesting demand response can be a useful tool to reduce the need for new power investment on the grid.
The Google scheme uses the existing ability to move loads to a different time or location, and simply gives an external power utility the power to request such an action.
"When we receive notice from a grid operator of a forecasted local grid event, for example an extreme weather event that will cause a supply constraint, we can alert our global computing planning system to when and where it will take place," says the blog. "This alert activates an algorithm that generates hour-by-hour instructions for specified data centers to limit non-urgent compute tasks for the duration of the grid event, and allows them to be rescheduled after the grid event has passed."
Some tasks will get rerouted to other data centers, while others are simply delayed. The jobs are chosen so as not to affect services including Search, Maps, YouTube, Google Cloud, and office applications like Gmail, the blog explains.
The system was tested in Europe, during winter 2022-23, when natural gas shortages put energy prices up. Google used the system to reduce energy demands at its facilities in the Netherlands, Belgium, Ireland, Finland, and Denmark during the typical peak period between 5pm and 9pm.
In the summers of 2022 and 2023, the system was used in Taiwan, to help out the country's isolated grid. With guidance from Taiwan Power Company, Google reduced its data center’s power consumption every day during peak hours.
Finally, in the US in Oregon and Nebraska, Google switched down data center energy demands during recent extreme weather events including heat waves and winter storms.
Following these successful pilots, Google says it is planning to use the technique more widely.
"Western US continues to improve system reliability, demand response capabilities will provide tremendous local and regional value during extreme events," said Chris Allen, cirector of Power Resources, Northern Wasco County PUD "Google and NWCPUD successfully demonstrated day-ahead DR capabilities at Google's facilities in the Dalles, Oregon under a recent pilot. We look forward to continuing to work with Google to expand and improve this flexibility as these capabilities mature."
In Europe, Lucien Wiegers, commercial director, Eneco Energy Trade said: "Our previous pilot in the Netherlands demonstrated Google’s responsive capability of data centers – and we’ll continue to work together to build on the promising results."
