Distributed Systems in the Age of AI: A Systematic Review of Sustainability Practices, Metrics, and Frameworks

Abstract

Today, cloud computing, edge intelligence, microservices, serverless computing, and federated artificial intelligence rely on distributed systems for computing resources. The growing need for sustainability studies is complicated by the intervening variables surrounding sustainability, which span such areas as software architecture, orchestration, network components, heterogeneous hardware, cooling, electricity generation, and hardware production. Moreover, AI presents a further complication to the sustainability discourse because it represents both a rapidly growing workload as well as a means of optimizing resource allocation, forecasting, scheduling, and controlling resources. This systematic integrative review covers findings until September 2026 related to sustainability in distributed systems. Reporting was guided by PRISMA 2020 and PRISMA-S principles for scope definition, concepts searched, and eligibility, extraction, interpretation, and synthesis processes, although a full PRISMA flow diagram could not be reconstructed because database export-level logs were unavailable. The final corpus included 96 sources, including over 80 peer-reviewed studies from published articles, standards, and measurement reports. This synthesis identifies five groups of best practices for the application/model layer, namely demand reduction; resource proportionality and consolidation; edge-cloud placement and workload shifting; carbon- and thermal-aware infrastructure management; and measurement/governance based on the principle of functional units and life-cycle boundaries. Apart from that, it has been discovered that energy is still the primary measure while operational carbon, embodied emissions, water, network effects, and hardware life cycle are still measured in less consistent manner. The frameworks currently available, including sustainability models, Sustainability Awareness Framework, Green Software Measurement Model, Software Carbon Intensity, and new green federated-learning frameworks address various parts of the problem but seldom provide a complete cross-layer measurement-and-control cycle at all. To fill in this gap, the review proposes SAIDS (Sustainability Aware AI for Distributed Systems) framework, which involves Measure-Attribute-Predict-Optimize-Verify-Report stages and relies on service, reliability, privacy, cost, and life-cycle dimensions.