期刊名称：Journal of Theoretical and Applied Information Technology
印刷版ISSN：1992-8645
电子版ISSN：1817-3195
出版年度：2015
卷号：80
期号：2
出版社：Journal of Theoretical and Applied
摘要：Data centers of today are rapidly moving towards the use of server virtualization as a preferred way of sharing a pool of server hardware resources between multiple �guest domains� that host different applications. The hypervisors of the virtualized servers, such as the Xen use fair schedulers to schedule the guest domains, according to priorities or weights assigned to the domain by administrators. The hosted application�s performance is sensitive to the scheduling parameters of the domain on which the application runs. However, the exact relationship between these parameters of the domain and the application performance measures such as response time or throughput is not obvious and not static as well. Furthermore, due to the dynamics present in the system there is need for continuous tuning of the scheduling parameters. The main contribution of our work is the design and implementation of a controller that optimizes the performance of applications running on guest domains. We focus on a scenario where a specific target for the response time of an application may not be provided. The goal is to dynamically compute the CPU shares for the virtual machines in such a way that the application throughput should be maximized, while keeping the response time as low as possible, with the minimum possible allocation of CPU share for the guest domain. The optimizing controller design is based on the feedback control theoretic concept. The controller computes the values of the scheduling parameters for every guest domain in such a way that it minimizes the CPU usage and response time, and maximizes throughput of the applications. To evaluate our work, we deployed multi-tier application in virtual machines hosted on the Xen virtual machine monitor. The performance evaluation results show that the controller brings the cap value close to the expected optimal value. The optimizing controller also rapidly responds to changes in the system when a disturbance task is introduced or load on the application is changed.
关键词：Cloud Computing; Data Center; Virtualization; hypervisor; Xen; virtual machine; Green IT; scheduler; performance; response time; throughput; feedback control theory.
