Thermal-Aware Optimization of Workload Distribution in Data Centers

Workload management has been increasingly critical and complex for sustainable data centers, integrating power management and thermal models. In this short paper, we present an optimization method for homogeneous server environments aiming at bringing the active servers inlet temperature closer to a desired data center-wide average value by means of workload allocation. We explore the mixed integer programming (MIP) approach to formulate our optimization problem. We claim that bringing servers´ temperature closer to the data center average one is an effective way of reducing hot spots, minimizing servers total power consumption while favoring the maximization of cooling temperature. The proposed optimization approach was validated with preliminary experiments using synthetic data.