Research On Resource Allocation Technology In Cloud Computing

With the rapid development of Internet and the growth of information, the requirement of large-scale computing and massive data processing in business applications, scientific computing and industry is increasing. Cloud computing emerging as a new business model and service model, it distributes tasks paralleled on data center consisted by large number of compute nodes. Cloud computing provides computing power, storage space, and information services on demand for various applications.The technology of efficient resources allocation is a key technology for cloud computing. The huge scale of resources and the use of virtualization technology, as well as real-time and variability of application requests brought many challenges to resources.allocation. According to resource allocation for cloud computing at different stages, this paper researches the key technologies of cloud computing resource allocation from three aspects of the resource provisioning, resource allocation, resource adjustment. The main content includes the following four aspects:(1) The resource provision method for Cloud service provider.According to the relationship of users, cloud service provider and cloud infrastructure provider, this paper proposes a two-stage resource allocation method for cloud service providers. At first, this paper focuses on the on-demand market and the spot market for real-time workload capacity planning, making out of the set of resources to optimize service providers profit. Because of the lack of measure about performance fluctuations as well as the customer’s satisfaction, this paper proposes a utility function to measure customer’s satisfaction. Then this paper designed an algorithm to optimize users’satisfaction. Finally, using the data of Amazon EC2in our experiment, the experimental results show that the proposed method effectively improves service providers’profit and gains better user’s satisfaction.(2) The virtual machine resource allocation method based on time and space constraints. Allocating virtual machine to physical server is a key problem in cloud computing environments. A multi-dimensional resource allocation algorithm based on time and space constraints is proposed. The algorithm calculates running time with different policy of virtual mechine placements according to the time constraint of virtual machine. Then the energy costs consumpted by physical machine is calculated. The energy comsuption and resources demand are spatial and temporal constraints of resource allocation. Aiming at minimize the resources used for allocate virtual machines, uses the improved particle swarm algorithm to solve the problem. The experimental results show that the proposed method effectively reduces the resource costs and improve data center resource utilization.(3) Resources adjustment method of server consolidation. During the application running, since the workload has the characteristics of periodic variation, so the data center resource needed to be adjusted to improve resource utilization when the condition of idle resources happens. This dissertation considers the cost of migration, communication and energy, and proposes a method of minimize the number of physical servers in order to optimize energy costs. Experimental results show that the proposed method can reduce the cost of migration and the total cost of communication between virtual machines as well as significantly reducing the number of working physical servers.(4) Elastic resources adjustment method for application performance. During the application running, there will be unpredictable load fluctuations. Therefore, it is necessary to design a reasonable adjustment method to ensure application QoS based on the optimization of resource utilization. This dissertation designs a fine-grained resource adjustment method, which is divided into three steps:the healing elastic resource adjustment method, physical resource layer resource adjustment method and the virtual machine layer resource adjustment method. The proposed methods monitor application response time and resource utilization:when the monitored value exceeds the threshold, it will trigger resource adjustment strategies at different levels of resources to scaling resource provided to application. Simulation results verify the effectiveness of the proposed method.