Integrated Control of Multiple Cooling Units

Abstract

Data centres are an integral part of today's technology. With the growing demand for data centers to meet computational needs, there is pressure to decrease data center-related costs. By reducing the amount of power needed to cool servers, the overall power consumption can be decreased. Efficient cooling of data centres involves meeting temperature constraints while minimizing power consumption. By exploring the opportunities that may be available through controlling multiple cooling units, we can avoid issues such as overcooling (some parts of the data center being cooled more than necessary) or warm air recirculation (return of exhausted hot air to inlets of servers). Currently, in data centres with more than one cooling unit, each of the cooling units is controlled independently. This mode of operation results in each cooling unit needing to be set for the worst case, which results in over cooling and is not energy efficient. Coordinating cooling units has the potential to decrease the power consumption of a data centre by eliminating this over cooling. Furthermore, coordinating with workload management may help mitigate cooling unit power consumption. This research is concerned with exploring what is feasible within the options discussed above. It contains two main parts. In the first part, we present an algorithm to minimize power consumption of cooling units while keeping all the server cores below a temperature threshold. In the second part of this thesis, we derived a data-driven model for server outlet temperature.