Distributed Virtual Machine Placement based on Dependability in Data Centers

One of primary and challenging difficulties in placing virtual machine (VM) onto physical machines in a cloud data center, is to balance the interests of both cloud service providers (CSPs) and cloud tenants, which are often expressed as minimizing the number of power-on servers and improving user experience respectively. However, the current research on VMP try to gathering the requested VMs on the fewest possible activated PMs for the purpose of reducing the number of power-on physical machines, without taking the experience of cloud tenants in to account. In this study, we define and formulate the VMP problem in cloud data center, and design a VMP strategy named Dependability based Distributed Virtual Machine Placement algorithm (D2VMP), with the placement goals of energy efficiency and dependability. To maintain the dependability of user requests D2VMP limits the number of virtual machines from the same request assigned on each physical machine. Furthermore, to ensure minimize the number of power-on physical machines, we propose to divide the data center into several clusters, and thus, different requests can reuse activated clusters. The experimental results demonstrate that proposed algorithm can effectively lower the power-on devices and maintain service-dependability.