GREEN ARTIFICIAL INTELLIGENCE FOR INTELLIGENT RESOURCE MANAGEMENT

Abstract

Artificial intelligence workloads and cloud services require increasingly sophisticated resource allocation, yet conventional scheduling policies may optimize utilization or response time without explicitly accounting for energy consumption and carbon intensity. This paper proposes a Green Artificial Intelligence framework for intelligent resource management that combines workload prediction, energy-aware optimization, service-level constraints, and runtime feedback. The framework estimates near-term demand from system telemetry and selects resource allocations that balance performance, energy use, and operational cost. A carbon-aware objective can incorporate time-varying grid carbon intensity where trustworthy data are available. The proposed evaluation uses trace-driven experiments and a reproducible comparison against static allocation, utilization-based autoscaling, and energy-unaware scheduling. Key measures include energy per completed task, average and tail latency, SLA violation rate, resource utilization, and estimated carbon emissions. A public cloud workload trace and a documented power-measurement model are recommended for implementation in Python. Because no implementation, trace processing, or experiments have been executed for this manuscript, all quantitative results are left pending.