On a quest to reduce the carbon emissions of your digital services, you might already have tried to automatically schedule the placement of your workloads at moments and in places where the carbon intensity of the grid is supposed to be lower. Namely, you might already have tried “carbon-aware computing”. While it seems a smart action to do, on paper, looking at the dynamics behind the scene reveals that the benefits of this practice are not so obvious, let alone that it actually isn’t producing the opposite effect targeted. In a study for the french ecological transition agency (ademe), we analyzed those effects in a consequential approach and will share the insights and results with you.
In this presentation, we’ll have a look at the methodology we used in the study, then the consequences tree built to detail the positive and negative dynamics identified as tied to carbon aware computing practices. Then we’ll go through the different branches of this tree, the consequences identified, give a bit of context and explaning the dynamic described. We’ll then summarize the insights coming from this analysis : depending on what carbon aware practice you are concerned with, whether it is workload shaping, workload time shifting or space shifting, what conditions are making you approach seemingly valid as a carbon reduction lever, or not, and why.