Fair Carbon Disaggregation and Scoped Attribution forCloud Applications

In this article, we motivate and show the feasibility of fine-grained metrics which can capture the various components of energy consumption and carbon emissions. The diversity of stakeholders, usecases, and optimization objectives requires finer-grained carbon metrics-going beyond the current coarse-grained scalar metrics such as SCI (Software Carbon Intensity). We identify two fundamental and central challenges in carbon footprint measurement: how the shared operational carbon is disaggregated among applications, and how embodied emissions can be scoped and attributed based on their locus of control. Different solutions to these challenges result in a spectrum of metrics. One such solution is the use of Shapley values, and we provide a concrete example of its application using serverless function carbon accounting.