Google shifting workloads to run when the sun will shine and the wind will blow

Already time-shifting inside some bit barns with electricity-source-predict-ometer

6 Reg comments Got Tips?

Google has revealed that it will try to shift work around its cloud to follow the availability of electricity generated by renewable resources.

The ads and search giant has discussed its “carbon-intelligent computing platform” that can “shift the timing of many compute tasks to when low-carbon power sources, like wind and solar, are most plentiful.”

This is not an entirely new idea: IBM won a patent for something rather similar-sounding in 2013 and the concept of “follow-the-moon” computing that moves workloads around the world to take advantage of off-peak power prices kicking in at night has been mooted for many years.

Google seems to be making this work by using two forecasts – one indicating future carbon intensity of the local electrical grid near its data centre and another of its own capacity requirements – and using that data “align compute tasks with times of low-carbon electricity supply.”

The result is that workloads run when Google thinks it can do so while generating the lowest-possible CO2 emissions.

Cisco rations VPNs for staff as strain of 100,000+ home workers hits its network

READ MORE

At present the scheme shifts workloads across a day and defers non-critical tasks like processing YouTube videos. But the company said its future plans aim to “shift load in both time and location”.

The company said it will publish the research in the hope others can use it to lessen the impact of their own cloudy workloads.

This tech has enormous potential. Researchers have already done work on predicting the price of AWS spot instances, so that buyers can find the cheapest-possible computing resources. Similar tools tuned to detect electricity pricing or sources could make it possible to send workloads to the cheapest and greenest sources. ®

SUBSCRIBE TO OUR WEEKLY TECH NEWSLETTER


Biting the hand that feeds IT © 1998–2020