Thermal-aware job scheduling in data centers:an optimization approach

Data centers are becoming more and more vital for every day life: they are huge computational beasts that run google queries, keep facebook profiles, and store our data in the cloud. As a result, research has focused the last decade on making the computing halls that host data centers more energy efficient. This thesis focuses on thermal management and control techniques that balance the thermal load in the data center, using knowledge of the thermal flows and leakages among the server equipment. For maximal utility, the controllers are extended to allow operation at the operating boundary. The results are completed by simulating a combination of thermal-aware controllers designed in this thesis, and power-aware controllers designed elsewhere. It is shown that a smart combination of both techniques results in greater energy reductions than implementing a single-purpose controller. Finally this thesis provides a method for characterizing the thermal map of any data center. With this identification method, it is possible to apply the controllers designed in this thesis to any existing data center