Energy-Efficient Green Computing Using AI-Based Workload Scheduling in Heterogeneous Cloud and Edge Computing Environments: A Comprehensive Review
Abstract
The rapid proliferation of data-intensive paradigms—including Internet of Things (IoT) ecosystems, autonomous vehicles, real-time healthcare telemetry, and generative artificial intelligence (AI)—has precipitated an exponential surge in global digital infrastructure energy consumption. Heterogeneous cloud-edge computing continuums have emerged as the foundational paradigm to process these ultra-scale, dynamic workloads by distributing tasks across centralized hyper-scale cloud datacenters and decentralized, localized edge processing nodes. However, managing energy efficiency across highly non-uniform processor architectures (e.g., CPUs, GPUs, TPUs, FPGAs) while satisfying strict, application-specific Quality of Service (QoS) guarantees remains a formidable scientific and engineering challenge. This review paper provides a rigorous, technical examination of state-of-the-art AI driven workload scheduling methodologies explicitly engineered for green computing across heterogeneous cloud-edge topologies. We synthesize foundational concepts of dynamic voltage and frequency scaling (DVFS), workload profiling, dynamic resource allocation, and AI optimization algorithms—spanning Deep Reinforcement Learning (DRL), Meta-Heuristics (GA, PSO, ACO), Federated Machine Learning, and Multi-Agent Reinforcement Learning (MARL). We present a granular domain architecture, formalize mathematical formulation models for multi-objective optimization (balancing Power Usage Effectiveness, carbon emission rate, response latencies, and execution cost), and critically evaluate state-of-the-art framework implementations. Furthermore, key empirical findings from published research, open research gaps, hardware-software co-design constraints, and practical industrial deployment guidelines are synthesized. This paper provides computer science researchers, systems architects, and infrastructure engineers with a comprehensive blueprint for designing next generation, self-adaptive, infrastructures. carbon-aware cloud-edge computing
KEYWORDS: Green Computing, AI-Based Workload Scheduling, Heterogeneous Cloud-Edge Continuum, Deep Reinforcement Learning, Multi Objective Optimization, Carbon-Aware Resource Allocation, Power Usage Effectiveness (PUE).
Full Text:
PDF 91-107Refbacks
- There are currently no refbacks.
Copyright 2017 ManTech Publications. All Rights Reserved