Automated management of virtualized data centers

Virtualized data centers enable sharing of resources among hosted applications. However, it is difficult to manage these data centers because of ever-changing application requirements. This thesis presents a collection of tools called AutoControl and LiteGreen, that automatically adapt to dynamic changes to achieve various SLOs (service level objectives) while maintaining high resource utilization, high application performance and low power consumption. AutoControl resource manager is based on control theory and optimization techniques. The resource manager is a combination of an online model estimator and a novel multi-input, multi-output (MEMO) resource controller. The model estimator captures the complex relationship between application performance and resource allocations, while the MIMO controller allocates the right amount of resources to achieve application SLOs. We also developed a power manager called LiteGreen to save desktop energy by virtualizing the users desktop computing environment as a virtual machine (VM) and then migrating it between the users physical desktop machine and a VM server, depending on whether the desktop computing environment is being actively used or is idle. AutoControl and LiteGreen are built using Xen and Hyper-V virtualization technologies and various experimental testbeds including a testbed on Emulab are built to evaluate different aspects of AutoControl and LiteGreen.