IaaS云环境下一种基于综合满意度的虚拟机放置策略

基础设施即服务（IaaS）环境下的一个关键需求是对租户申请的虚拟机进行合理放置。当前虚拟机放置策略的研究大都集中在数据中心能耗、资源损耗以及负载均衡等方面，很少有工作关注其对租户虚拟机启动时间的影响。为了减少虚拟机请求的周转时间，降低数据中心的资源损耗，本文首先建立了云服务租户满意度模型，给出了虚拟机请求到达云端后周转时间的量化方法；然后基于数据中心的资源损耗建立了云服务提供商满意度模型；最后，基于租户虚拟机启动时间与系统资源损耗建立了多目标约束优化模型，并提出了一种基于综合满意度（Comprehensive Satisfaction Based，CS-B）的虚拟机放置策略，该策略综合考虑了云服务租户与云服务提供商的需求，将租户所申请的虚拟机放置到综合满意度最高的服务器中运行。在OpenStack云平台上的仿真实验表明，CS-B虚拟机放置策略能够有效减少租户虚拟机在云端的部署时间，降低数据中心的资源损耗，有效提高了云服务商及租户的满意度。

A Comprehensive Satisfaction Based Virtual Machine Placement Policy in IaaS Cloud

The key requirement of the infrastructure as a service (IaaS) environment is to properly arrange the virtual machine applied by tenants. Most of the current researches on virtual machine placement strategy focus on data center energy consumption, resource loss and load balancing. Little attention has been paid to the impact of virtual machine placement on renter''s virtual machine startup time. Virtual machine startup time is too slow, which will cause the current load of tenants not to be allocated to other virtual machines in time, thus hindering the horizontal expansion of applications. In order to reduce the turn-around time of virtual machine request and the resource loss of data center, we establish a cloud service tenant satisfaction model based priority of tenant request, and give a quantitative measurement of time cost of deploying at first. Then, the cloud service provider satisfaction model is established according to the resource loss of the data center. Finally, a multi-objective constraint optimization model based on the starting time of the tenant virtual machine and the loss of the system resources, and a virtual machine placement policy is proposed based on comprehensive satisfaction (CS-B) which takes into account the needs of cloud services tenants and cloud service providers by giving different weights, then place the virtual machine that the tenant applies to the server with the highest level of comprehensive satisfaction. Compared with other algorithms, the proposed algorithm has the characteristics of low complexity, simple parameters and easy to combine with the real cloud platform (such as OpenStack, CloudStack, Eucalyptus). The experiment results under the OpenStack cloud platform show that it can effectively reduce the deployment time of the virtual machine, and reduce the resource loss of the data center, and then increase the satisfaction of the cloud service tenant and the cloud service provider.