Joint Optimization of Task Scheduling and Thermal Management for Geo-Distributed Data Centers

The growing demand for artificial intelligence (AI) workloads is significantly increasing data center energy consumption. This paper presents D3T, a dual-timescale optimization framework for task scheduling and thermal management in geo-distributed data centers. D3T integrates task scheduling with dynamic cooling control to address the mismatch in response times between IT and cooling subsystems. Using Deep Q-Networks (DQN) for fast task scheduling and QMIX for thermal management, D3T reduces energy consumption and operational costs. Experimental results demonstrate up to 29% reduction in cooling OPEX and 18% in IT OPEX, while maintaining service quality,