The role of energy consumption prediction in Green Orchestration

Open Webpage

Ioannis Korontanis, Ioannis Kontopoulos, Konstantinos Tserpes, Iraklis Varlamis

Published: 15 Dec 2025, Last Modified: 05 May 2026CrossrefEveryoneRevisionsCC BY-SA 4.0
Abstract: This paper introduces a specialized profiler for energy consumption prediction and proposes a Conceptual Green Orchestration Pipeline that utilizes the profiler as middleware. The system's architecture integrates a standard orchestrator (like Kubernetes) with a local ONNX model registry, targeting edge devices for deployment. Crucially, the profiler demonstrated strong performance, achieving 89% accuracy in selecting the optimal deployment device for resource-intensive models, a significant improvement over the 33.4% accuracy seen with random placement.