Load Flexibility Modeling of CPU-Intensive Computing Tasks in Data Centers

Current data centers possess significant, yet underutilized, potential for load flexibility. One key limitation is the lack of fine-grained modeling of their energy consumption characteristics. In this paper, we focus on CPU-intensive computing tasks and propose a quantitative flexibility modeling framework by analyzing the impact of energy sensitivity parametersspecifically, thread count and CPU frequency-on power consumption, energy usage, and task duration. We investigate energy consumption characteristics in both single-server and multiserver environments and examine the effects of task-level and joblevel allocation strategies. Experimental validation demonstrates that our models achieve over 85% prediction accuracy and show that CPU-intensive computing tasks can offer up to 30% power modulation. These results support the use of parameter-level tuning for flexible and energy-aware scheduling in data centers.