International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 13 Issue: 02 | Feb 2026
p-ISSN: 2395-0072
www.irjet.net
A REVIEW OF CARBON-AWARE INTELLIGENT SCHEDULING OF MACHINE LEARNING WORKLOADS IN GEOGRAPHICALLY DISTRIBUTED DATA CENTERS Sachin Kumar Gupta1, Mrs. Arifa Khan2 1Master of Technology, Computer Science and Engineering, Lucknow Institute of Technology, Lucknow, India 2Assistant Professor, Department of Computer Science and Engineering, Lucknow Institute of Technology,
Lucknow, India ---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - The rapid expansion of machine learning (ML)
and carbon emissions. Data centers, which serve as the backbone of digital infrastructure, are now recognized as significant energy consumers worldwide. Within this context, carbon-aware intelligent scheduling of machine learning (ML) workloads has emerged as a promising strategy to balance computational performance with environmental sustainability. This section introduces the background, motivation, and scope of this review.
applications has significantly increased the computational demand of modern data centers, intensifying their energy consumption and associated carbon emissions. As geographically distributed data centers become the backbone of large-scale AI services, carbon-aware intelligent scheduling has emerged as a promising strategy to mitigate environmental impact while maintaining performance guarantees. This review systematically examines existing approaches to carbon-aware scheduling of ML workloads across geographically distributed infrastructures. We analyze methods that leverage real-time and forecasted grid carbon intensity, renewable energy availability, geographic load balancing, and intelligent optimization techniques such as machine learning–driven prediction and reinforcement learning–based decision making. The review categorizes prior studies based on scheduling objectives, optimization strategies, and evaluation metrics, highlighting trade-offs among carbon reduction, latency, cost, and quality of service. Furthermore, we compare architectural frameworks, workload characteristics, and deployment assumptions to identify common design patterns and limitations. Key challenges—including data uncertainty, scalability constraints, multi-objective conflicts, and lack of standardized benchmarks—are critically discussed. Finally, emerging research directions are outlined, emphasizing integrated AIdriven orchestration, cross-layer optimization, and sustainability-aware system design. By synthesizing current knowledge and identifying research gaps, this review aims to provide a comprehensive reference for researchers and practitioners seeking to design environmentally sustainable ML scheduling strategies in distributed cloud environments.
1.1 Background 1.1.1 Climate Change and Carbon Emissions from Digital Infrastructure Global climate change is strongly linked to greenhouse gas emissions, particularly carbon dioxide (CO₂), generated from fossil fuel–based energy systems. The information and communication technology (ICT) sector contributes a nonnegligible share of global emissions, driven by increasing internet usage, cloud computing, and AI applications. Recent assessments indicate that digital infrastructure, including networks and data centers, accounts for a growing percentage of global electricity consumption (IEA, 2023). Although improvements in hardware efficiency have partially mitigated growth in energy demand, the overall carbon footprint of computing continues to rise due to expanding service requirements. Digital platforms increasingly rely on high-performance computing clusters to support data-intensive tasks such as deep learning model training and large-scale analytics. Studies have shown that training large neural networks can produce substantial carbon emissions, especially when powered by carbon-intensive grids (Strubell, Ganesh and McCallum, 2019). Consequently, sustainability has become a critical design objective in next-generation computing systems.
Key Words: Carbon-aware computing; Machine learning workloads; Geographically distributed data centers; Intelligent scheduling; Renewable energy integration; Sustainable cloud computing.
1.1.2 Role of Data Canters in Global Energy Consumption
1. INTRODUCTION
Data centers form the core infrastructure supporting cloud services, enterprise applications, and AI systems. Their energy consumption arises from computing equipment, storage systems, networking devices, and cooling infrastructure. According to global energy assessments, data
The rapid digital transformation of modern society has led to an unprecedented expansion of cloud services, artificial intelligence (AI), and large-scale data processing systems. While these advancements drive innovation and economic growth, they also contribute to increasing energy demand
© 2026, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 855