Date of Award
Doctor of Philosophy
Electrical and Computer Engineering
J. Morris Chang
In recent years, cloud computing services continue to grow and has become more pervasive and indispensable in people's lives. The energy consumption continues to rise as more and more data centers are being built. How to provide a more energy efficient data center infrastructure that can support today's cloud computing services has become one of the most important issues in the field of cloud computing research.
In this thesis, we mainly tackle three research problems: 1. how to achieve energy savings in a virtualized data center environment; 2. how to maintain service level agreements; 3. how to make our design practical for actual implementation in enterprise data centers. Combining all the studies above, we propose an optimization framework named CoolCloud to minimize energy consumption in virtualized data centers with the service level agreement taken into consideration. The proposed framework minimizes energy at two different layers: (1) minimize local server energy using dynamic voltage \& frequency scaling (DVFS) exploiting runtime program phases. (2) minimize global cluster energy using dynamic mapping between virtual machines (VMs) and servers based on each VM's resource requirement. Such optimization leads to the most economical way to operate an enterprise data center.
On each local server, we develop a voltage and frequency scheduler that can provide CPU energy savings under applications' or virtual machines' specified SLA requirements by exploiting applications' run-time program phases. At the cluster level, we propose a practical solution for managing the mappings of VMs to physical servers. This framework solves the problem of finding the most energy efficient way (least resource wastage and least power consumption) of placing the VMs considering their resource requirements.
Zhang, Zhiming, "CoolCloud: improving energy efficiency in virtualized data centers" (2016). Graduate Theses and Dissertations. 15213.