LAWA: Loss Aware Workload Assignment in Data Centers

Abstract

Data centers are the backbone of computing, storing, processing and allowing access to large amounts of data. They consume a huge amount of energy, as of 2018, each year consuming approximately 200 TWh and this amount is only going to increase. A lot of work has been done to increase the energy efficiency of data centers. Research on reducing the power consumption of data centers has examined different levels of operations. Some of this research has proposed algorithms in lower levels such as at the device level and some in higher levels such as servers, cooling systems and workload managers. But there is no work that has considered inefficiencies in power distribution to servers (server power loss). This thesis proposes an algorithm that minimizes server power loss and then compares the resulting algorithms to those that do not consider server power loss to determine whether the resulting algorithms are consistent or not. It will be shown that the power savings of such an algorithm can be quite significant and as a result it is crucial to include such considerations when performing workload assignment. Finally, some opportunities to continue this work in the future will be discussed.