Google has revealed details of a carbon-reduction trial running within its datacentres that ensures non-urgent compute tasks are undertaken when its facilities are most likely to be powered by renewable sources.
This is made possible through the deployment of an in-house developed carbon-intelligent computing system within the search giant’s hyperscale datacentres that uses forecast data to predict when supplies of wind and solar power are likely to be at their most plentiful.
The system then ensures that non-urgent tasks, such as adding new words to Google Translate or the processing of YouTube videos, are carried out during these peak renewable energy supply periods.
“This is done without additional hardware and without impacting on the performance of Google services like Search, Maps and YouTube that people rely on around the clock,” said Ana Radovanovic, technical lead for the Carbon-Intelligent Computing system, in a blog post.
“Shifting the timing of non-urgent compute tasks – like creating new filter features on Google Photos, YouTube video process or adding new words to Google Translate – helps reduce the electrical grid’s carbon footprint, getting us closer to 24×7 carbon-free energy.”
This is a sustainability goal that Google has now set its sights on achieving, having previously succeeded at becoming a carbon neutral company in 2007, and ensuring – for the past three years – that the energy consumed by its operations is matched by renewable energy purchases.
“Now, we’re working toward 24×7 carbon-free energy everywhere we have datacentres, which deliver our products to billions of people around the world. To achieve 24×7 carbon-free energy, our datacentres need to work more closely with carbon-free energy sources like solar and wind,” Radovanovic wrote.
Ana Radovanovic, Google
The carbon-intelligent platform is the creation of a small engineering team within Google, she said in the blog post, and is being used within every datacentre the company operates.
It works by using forecast data supplied by a third-party provider called Tomorrow that predicts the average hourly carbon intensity of the electrical grid that supplies a particular datacentre and how that is set to change over the course of the day.
This is then combined with Google’s own internal forecast data that predicts the hourly power resources the same datacentre is likely to need to carry out its compute tasks during the same period of time.
“We use the two forecasts to optimise hour-by-hour guidelines to align compute tasks with times of low-carbon electricity supply,” said Radovanovic. “Results from our pilot suggest that by shifting compute jobs we can increase the amount of lower-carbon energy we consume.”
Based on this finding, the team is now looking to scale up its ambitions with regard to boosting the flexibility of the tasks that can be scheduled in this way, but also the location of where they are completed.
“The first version of this carbon-intelligent computing platform focuses on shifting tasks to different times of the day, within the same datacentre. But, it’s also possible to move flexible compute tasks between different datacentres, so that more work is completed when and where doing so is more environmentally friendly,” the blog post continued.
“Our plan for the future is to shift load in both time and location to maximise the reduction in grid-level CO2 emissions.”
And beyond that, the firm hopes its learnings will inspire other datacentre operators to follow its lead and adopt similar processes and procedures in their own facilities.
“Our methodology, including performance results of our global roll-out, will be shared in upcoming research publications. We hope that our findings inspire other organisations to deploy their own versions of a carbon-intelligent platform, and together, we can continue to encourage the growth of carbon-free electricity worldwide,” Radovanovic added.
Aside from Google’s ongoing efforts to decarbonise its operations, the carbon-free intelligence platform is the latest in a string of innovations at datacentre level the firm has introduced in recent years to make its server farms more energy efficient and environmentally friendly.
As previously documented by Computer Weekly, the firm has previously made use of machine learning and artificial intelligence technologies to optimise the cooling of its datacentres to, in turn, reduce the amount of energy they consume.
That system has been credited in the past with helping Google to cut the energy consumption of its datacentres by 30%.