From Measurement to Emissions: Assessing the Carbon Footprint of Traffic Flows

As sustainability becomes a requirement in network operations, accurately quantifying the carbon footprint of Internet traffic is essential. While energy-aware networking has seen significant attention, the ability to trace carbon emissions at the flow level remains an open challenge due to the complexity of shared infrastructure and lack of related telemetry. In this paper, we present a methodology to obtain fine-grained per-flow carbon emissions from traffic statistics. To this end, we collect power measurements from three switches under varying traffic conditions, including synthetic and real-world traces. From these measurements, we derive a regression model that accurately estimates instantaneous router power consumption using only throughput and packet rate counters--achieving >96% accuracy across all switch types and traces. We then extend this model to compute per-flow carbon emissions, distinguishing between consequential and attributional perspectives, and validate the results using traces from CAIDA and Google services. Our findings uncover actionable insights into how flow and network characteristics such as packet size, packet rate, and network utilization influence carbon cost. Finally, we propose feasible deployment strategies for flow-level carbon estimation frameworks. This work provides a foundational step towards enabling carbon-aware flow-level decision-making for users, applications, and network operators.