Performance And Energy Efficiency Of Virtual Machine Based Clouds

Combining high-speed network accesses and powerful computer virtualization, cloud computing provides an elastic and cost-effective service paradigm for the development of resource-hungry new applications as well as the expansion of a broad spectrum of existing applications. Virtualization has become wide spread in modern computers and operating systems; the extensive use of virtual machines (VMs) in a networked cloud environment, however, comes with unprecedented overhead. In this thesis we study the impact of virtualization overhead on real-world systems. First, we present a study on the performance of modern virtualization solutions under DoS attacks. We experiment with the full spectrum of modern virtualization techniques, from paravirtualization, hardware virtualization, to container virtualization, with a comprehensive set of benchmarks. Our results reveal severe vulnerability of modern virtualization. Further, this thesis presents an empirical study on the power consumption of typical virtualization packages while performing network tasks. We find that both Hardware Virtualization and Paravirtualization add considerable energy overhead, affecting both sending and receiving, and a busy virtualized web-server may consume 40% more energy than its non-virtualized counterparts. Next, we focused on cloud network performance instability in the public cloud. Through measurement of real world cloud platforms, we find that network performance degradation and variation phenomena can be prevalent and significant with both TCP and UDP traffic, even within the same data center, and even with a lightly utilized network. Our in-depth measurement and detailed system analysis reveal that the performance variation and degradation are mainly due to the requirements of the CPU in both computation and network communication. Finally, we study the use of virtual machine based clouds to provide infrastructure to support cloud gaming. For each issue we further suggest solutions to resolve performance issues associated with virtualized environments, both at the hypervisor level and, inside a VM, at the operating system level. Our implementation on both large public clouds and our cloud testbed have demonstrated their practicality as well as their effectiveness in improving and stabilizing the energy consumption and performance of virtual machine based clouds.