Energy management for virtual machines.

详细信息   

• 作者：Ye ; Lei.
• 学历：Doctor
• 年：2013
• 导师：Gniady,Christopher,eadvisorGniady,Christopherecommittee memberHartman,John H.ecommittee memberLowenthal,David K.ecommittee memberZhang,Beichuanecommittee member
• 毕业院校：University of Arizona
• Department：Computer Science.
• ISBN：9781303019647
• CBH：3557957
• Country：USA
• 语种：English
• FileSize：612499
• Pages：103

文摘

Current computing infrastructures use virtualization to increase resource utilization by deploying multiple virtual machines on the same hardware. Virtualization is particularly attractive for data center,cloud computing,and hosting services； in these environments computer systems are typically configured to have fast processors,large physical memory and huge storage capable of supporting concurrent execution of virtual machines. Subsequently,this high demand for resources is directly translating into higher energy consumption and monetary costs. Increasingly managing energy consumption of virtual machines is becoming critical. However,virtual machines make the energy management more challenging because a layer of virtualization separates hardware from the guest operating system executing inside a virtual machine. This dissertation addresses the challenge of designing energy-efficient storage,memory and buffer cache for virtual machines by exploring innovative mechanisms as well as existing approaches. We analyze the architecture of an open-source virtual machine platform Xen and address energy management on each subsystem. For storage system,we study the I/O behavior of the virtual machine systems. We address the isolation between virtual machine monitor and virtual machines,and increase the burstiness of disk accesses to improve energy efficiency. In addition,we propose a transparent energy management on main memory for any types of guest operating systems running inside virtual machines. Furthermore,we design a dedicated mechanism for the buffer cache based on the fact that data-intensive applications heavily rely on a large buffer cache that occupies a majority of physical memory. We also propose a novel hybrid mechanism that is able to improve energy efficiency for any memory access. All the mechanisms achieve significant energy savings while lowering the impact on performance for virtual machines.