CarbonEdge: Carbon-Aware Deep Learning Inference Framework for Sustainable Edge Computing

2026-03-28

AI Summary

This work addresses the high carbon emissions associated with edge AI inference—a critical yet often overlooked environmental challenge in edge computing. To tackle this issue, the authors propose CarbonEdge, a novel framework that, for the first time, explicitly incorporates carbon efficiency into the scheduling objective. CarbonEdge achieves a tunable trade-off between performance and carbon footprint in heterogeneous edge environments through adaptive model partitioning, real-time carbon footprint estimation, and a carbon-efficiency-weighted scheduling algorithm. Experimental results demonstrate that in its CarbonEdge-Green mode, the framework reduces carbon emissions by 22.9% compared to monolithic execution, improves carbon efficiency by 1.3×—reaching 245.8 inferences per gram of CO₂—and incurs only 0.03 ms of scheduling overhead per task.

Abstract

Deep learning applications at the network edge lead to a significant growth in AI-related carbon emissions, presenting a critical sustainability challenge. The existing edge computing frameworks optimize for latency and throughput, but they largely ignore the environmental impact of inference workloads. This paper introduces CarbonEdge, a carbon-aware deep learning inference framework that extends adaptive model partitioning with carbon footprint estimation and green scheduling apabilities. We propose a carbon-aware scheduling algorithm that extends traditional weighted scoring with a carbon efficiency metric, supporting a tunable performance--carbon trade-off (demonstrated via weight sweep). Experimental evaluations on Docker-simulated heterogeneous edge environments show that CarbonEdge-Green mode achieves a 22.9% reduction in carbon emissions compared to monolithic execution. The framework achieves 1.3x improvement in carbon efficiency (245.8 vs 189.5 inferences per gram CO2) with negligible scheduling overhead (0.03ms per task). These results highlight the framework's potential for sustainable edge AI deployment, providing researchers and practitioners a tool to quantify and minimize the environmental footprint of distributed deep learning inference.

Problem

• carbon emissions
• edge computing
• deep learning inference
• sustainability
• environmental impact

Innovation

• carbon-aware scheduling
• adaptive model partitioning
• edge AI sustainability
• carbon efficiency
• green edge computing