获益驱动的虚拟机资源动态管理策略研究

发布时间：2018-08-20 11:26

【摘要】：作为一种新兴的商业模式,云计算通过开放的技术和标准把软硬件虚拟化成动态的资源,并以按需服务的形式提供给用户。云计算不仅能够大幅降低购买硬件、能源、制冷等成本,还能够加快应用程序的部署。云计算的核心特征为按需服务,而这也使得云计算环境下资源的管理成为一大难题。由于云资源提供商和云资源使用者之间存在利益上的冲突,现有的大多数研究都是从单一角色出发,研究如何通过资源管理使单一角色受益。在某种程度上,云资源提供商和云资源使用者之间是利益共同体。如果没有云资源使用者租用资源,云资源提供商就不能获取利润；而如果没有云资源提供商提供资源,云资源使用者就不能降低成本。因此,在进行资源管理时,必须综合考虑云资源提供商和云资源使用者的获益,在提高云资源使用者满意度的前提下,实现云资源提供商利润的最大化。首先,研究客户满意度驱动的虚拟机资源分配策略,从合理定价和性能保障两方面来提高云资源使用者的满意度。在定价方面,引入了微观经济学中的效用理论；在性能保障方面,使用负载预测和基于排队论的性能预测模型。针对现有的负载预测方法只适用于单层云服务,或只考虑请求总量而忽略其他因素造成预测不准的问题,提出了多因素感知的负载预测模型(MAPM),综合考虑请求总量、服务时间和最终用户偏好对负载的影响,以提高预测准确度。接着,使用排队论对多层云服务进行性能建模,计算出客户所需资源,为资源调度打下基础。实验证明,相对于只考虑请求总量的负载预测算法,多因素感知的负载预测模型(MAPM)能提高预测的准确度。然后,研究利润驱动的虚拟机资源调度策略,最大化云资源提供商的利润。首先,建立云资源提供商的利润模型,并根据微观经济学的原理,计算出能够使云资源提供商利润最大化的虚拟机资源数目。接着,以粒子群算法为指导,对数据中心虚拟机资源的调度进行建模分析,根据客户需求和当前数据中心的负载状态,以最大化资源利用率、最小化所占用物理机的数目及虚拟机迁移次数三个方面为目标,提出了基于增强型多目标粒子群算法的VM资源调度策略(EPSO-VM)来实现虚拟机资源的优化调度。实验证明,相对于标准粒子群算法以及贪婪算法,该策略能提高资源的利用率和云资源提供商的利润,同时降低占用物理机的数目和虚拟机迁移次数。
[Abstract]:As a new business model, cloud computing virtualizes software and hardware into dynamic resources through open technologies and standards and provides them with on-demand services. Cloud computing not only significantly reduces the cost of buying hardware, energy, and refrigeration, but also speeds up the deployment of applications. The core feature of cloud computing is on demand service, which makes resource management in cloud computing environment a big problem. Due to the conflict of interests between cloud resource providers and cloud resource users, most of the existing studies focus on how to benefit a single role through resource management. To some extent, there is a community of interests between cloud resource providers and cloud resource users. If there are no cloud resource users renting resources, cloud resource providers can not obtain profits, and if cloud resource providers do not provide resources, cloud resource consumers can not reduce costs. Therefore, in resource management, the benefits of cloud resource providers and cloud resource users must be considered comprehensively, and the profit of cloud resource providers should be maximized on the premise of increasing the satisfaction of cloud resource users. Firstly, the strategy of virtual machine resource allocation driven by customer satisfaction is studied to improve the satisfaction of cloud resource users from two aspects: reasonable pricing and performance guarantee. In the aspect of pricing, the utility theory of microeconomics is introduced, and the performance prediction model based on queuing theory and load forecasting is used in performance assurance. To solve the problem that the existing load prediction methods are only suitable for single-layer cloud services or only consider the total number of requests and ignore other factors, a multi-factor perceived load prediction model (MAPM),) is proposed to comprehensively consider the total amount of requests. The effect of service time and end-user preferences on the load to improve prediction accuracy. Then, the queuing theory is used to model the performance of multi-layer cloud services, and the resources required by customers are calculated, which lays the foundation for resource scheduling. Experimental results show that the multi-factor perceived load prediction model (MAPM) can improve the prediction accuracy compared with the load prediction algorithm which only considers the total amount of requests. Then, the profit driven virtual machine resource scheduling strategy is studied to maximize the profit of cloud resource providers. Firstly, the profit model of cloud resource provider is established, and according to the principle of microeconomics, the number of virtual machine resources that can maximize the profit of cloud resource provider is calculated. Then, under the guidance of particle swarm optimization algorithm, the scheduling of virtual machine resources in data center is modeled and analyzed. According to the customer demand and the current load state of the data center, the resource utilization can be maximized. Aiming at minimizing the number of physical machines occupied and the number of virtual machines migrated, a VM resource scheduling strategy (EPSO-VM) based on enhanced multi-objective particle swarm optimization algorithm is proposed to realize the optimal scheduling of virtual machine resources. Experiments show that compared with the standard particle swarm optimization and greedy algorithm, this strategy can improve the utilization of resources and the profit of cloud resource providers, and reduce the number of physical machines and the number of virtual machine migration.
【学位授予单位】：广西大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP302

【相似文献】