A cluster-oriented task assignment optimization for green high-performance computing center operations

The escalating demand for high-performance computing (HPC) centers necessitates efficient energy management. Prior research predominantly concentrated on individual servers, neglecting the workload interactions in resource sharing and task scaling environments. This study bridges this gap by developing an energy-aware task assignment model that integrates rack selection, task scaling, and resource allocation. Leveraging real-world data, we adopt a data-driven approach to manage the workload of a rack with multiple servers. We first identify pivotal workload characteristics affecting power consumption and investigate trade-offs between computing and luqid-cooling systems. We then propose a deep learning model to capture latent workload interactions further. We also design an adaptive large neighborhood search-based algorithm for task assignment optimizations. Simulations validate the performance of our models and highlight the impact of task queues on energy conservation. This study provides a holistic framework and managerial implications for energy efficiency management in HPC centers, integrating interpretive, predictive, and prescriptive analytics. © 2025 Elsevier Ltd.