Carbon-Aware Resource Management in Cloud, Edge, and AI Platforms: A Comprehensive Survey

Abstract

The rapid expansion of cloud computing, big data analytics, Internet-of-Things (IoT) platforms, and artificial intelligence has significantly increased the environmental impact of modern computing systems. While numerous studies propose techniques to improve energy efficiency, existing research remains fragmented, focusing on individual components such as data centers, scheduling algorithms, or machine learning models. This paper presents a comprehensive survey of sustainable computing approaches through a layered analysis of carbon-reduction strategies across the computing ecosystem. The reviewed literature is categorized into six operational layers: carbon measurement and accounting, infrastructure optimization, carbon-aware workload scheduling, sustainable data processing, green artificial intelligence, and continuous application services. The survey analyzes the effectiveness, trade-offs, and validation methods of existing techniques and highlights key limitations, including unreliable emission measurement, simulation-based evaluation, and conflicts between performance and sustainability. Results show that isolated optimizations often shift emissions rather than reduce total environmental impact. The study further identifies major research challenges such as lack of standardized carbon metrics, neglect of hardware lifecycle emissions, and absence of cross-layer coordination. Finally, future directions are outlined, emphasizing carbon-aware orchestration, lifecycle-aware system design, and renewable-integrated computing. The findings demonstrate that meaningful carbon reduction requires coordinated system-level optimization rather than isolated improvements within individual computing components.