Estimating the Carbon Footprint of Serverless Functions on a Public Cloud Platform

As the carbon footprint of cloud data centers grows rapidly, sustainability has become an increasing concern for practitioners. Understanding the carbon emissions of cloud workloads and identifying strategies to reduce them is critical. In this paper, we model and extensively analyze the carbon emissions of functions executed on a public serverless platform using available telemetry, offering new insights into the relationship between carbon emissions and traditional metrics of cost and performance. We explore various factors affecting carbon emissions, including host region, architecture, cold starts, application resource composition, and input-sensitivity. Based on our findings, we propose future optimization opportunities and research directions. Our work aims to empower developers to make more sustainable decisions when configuring or optimizing their applications.