Models and Control Strategies for Data Center Energy Efficiency

As the foundation of the nation’s information infrastructure, data centers have been growing rapidly in both number and capacity to meet the increasing demands for highlyresponsive computing and massive storage. Data center energy consumption doubled from 2000 to 2006, reaching a value of 60 TWh/year (Tera Watt hour / year). Coupled with increasing power and cooling demands imposed by the Moore’s law and with the quest for high density data centers, this trend has been rapidly raising the energy cost associated with data centers. Data centers are large cyber-physical systems (CPSs) with hundreds of variables that can be measured and controlled. Dynamics of the controlled processes span multiple time scales: electricity costs can fluctuate hourly, temperatures evolve in the order of minutes, and CPU power states can be changed as frequent as milliseconds. Processes also differ in the spatial areas they influence: computer room air conditioners (CRAC) affect the inlet air temperatures of multiple servers, whereas CPU power states affect only single servers. The large number of constraints and their heterogeneity in nature make data center control a challenging research problem. This dissertation considers data centers as CPSs, with a focus on run-time management and operating costs. The proposed modeling framework explicitly captures the cyberphysical nature of data centers and allows the development of models that represent both the computational and the thermal characteristics of a data center, as well as their interactions. The proposed control strategy attempts to manage both the computational and the thermal characteristics of a data center. The control strategy is based on a hierarchical/distributed control architecture that takes advantage of the modularity typically found in data centers. The hierarchy constitutes of three control levels. The lower levels of the hierarchy deal with fast dynamic processes, while the higher levels deal with the bulk thermal management and the coordination of the controllers at the lower levels. The focus