Research On Strategy Of Benefit-Driven Virtual Machine Resource Management

As a new business paradigm, cloud computing can virtualize hardware and software resources into dynamic resources through open technologies and standards, and then provide them to users in an on-demand manner, which makes it possible for computing resources to be the fifth public utility. Cloud computing can significantly reduce the cost of hardware purchasing, energy, cooling, and speed up the deployment of applications. But the core feature of cloud computing, service on-demand, makes resource management a major problem under the computing environment. And because of the interest conflicts between the Infrastructures Providers and Service Providers, current studies mainly focus on one party and how to make the single party benefit. To some extent, however, the Infrastructures Providers and Service Providers share the same interests. That is to say, the Infrastructures Providers can’t make profits if the Service Providers stop renting resources. Likewise, the Service Providers can’t reduce costs without Infrastructures Providers providing resources. Therefore, resource management strategies have to benefit both parties, maximizing the profits of the Infrastructure Providers and meeting the needs of Service Providers.First of all, a customer satisfaction-driven resource management strategy is proposed to improve user satisfaction in two ways, which is reasonable pricing and performance guarantee. In terms of pricing, the utility theory of microeconomics is introduced, while in performance guarantee, workload prediction and a performance model based on queuing theory are conducted. Existing workload prediction methods, either tailor-made for single-layer cloud services, or taking only the request volume into consideration, result in predicting inaccurately. Therefore, a Multi-Factors Aware Workload Prediction Model(MAPM) is proposed, taking consideration of request volume, service time, and end user preference, to improve the prediction precision. Then, using the queuing theory to model multi-tier cloud services and calculate demanded resources, which lay the foundation for resources provision. Experiments show that compared to the traditional request volume-based workload prediction, our algorithm can improve prediction precision.Then, a profit-driven virtual machine resource allocation strategy is introduced to maximize the profits of Infrastructures Providers. First, an Infrastructure Providers’profit model is created, and based on the principle of microeconomics, the number of virtual machines which can maximize the profit of Infrastructures Providers can be calculated. Then, using the particle swarm optimization theory as a guide, the virtual machine resource scheduling in the data center is modeled and analyzed. Based on the demands of users and the current workload of the data center, and aiming at maximizing resource utilization, minimizing the number of occupied physical machines and virtual machine migrations, an Enhanced Multi-Objectives Particle Swarm Optimization based Virtual Machine Scheduling Strategy(EPSO-VM) is adopted to realize optimization scheduling. Compared to the standard particle swarm algorithm and greedy algorithm, experiments show EPSO-VM can improve resource utilization and profit of Infrastructure Providers, and at the same time, reduce the number of occupied physical machines and virtual machine migrations.