A dynamic stochastic optimization control model for data centers based on numerical modeling

Energy optimization of Cooling Room Air Conditioners (CRACs) in data centers remains a key concern due to the increasing energy consumption of air-cooled blade servers in the space limits imposed by chassis and racks. For this issue, server workload is either randomly dispersed or equalized to avoid placing intensive workloads on specific servers. Computational Fluid Dynamics has been used but few papers report the impact of time-varying thermodynamics in the optimization control framework, i.e., most assume steady air flow. This paper proposes a dynamic stochastic optimization control method for enhancing CRAC energy efficiency by dynamically changing the workloads of both overall servers and CRACs. Two new constraints are assumed; first, temperature changes of nearby server outlets can lead to efficient cooling. Second, server inlet temperatures depend on the distance between servers and CRACs. Tikhonov regularization theory is introduced to handle such constraints. In numerical simulations of over 1500 different combinations, unlike existing reports that examine just a few combinations, our proposed optimization method is shown to be very effective in reducing the energy cost of CRACs compared with evenly loading the servers.