A Study Of Virtual Resources Scheduling Algorithm Based On Combinatorial Optimization In Cloud Computing

The cloud computing technology improves the resource utilization of a cluster by integrating and redistributing the physical resources. Virtualization technology is the foundation of the physical resources’ abstract, classification and integration and the capacity of providing elastic computing services. In the hierarchical model of cloud computing, infrastructure as a service(Iaa S) completed the process of abstraction of physical resources, all resource requests can be obtained from the resource pool. In this way, cloud computing platform can manage the computing cluster consisting of heterogeneous physical machines. In this scenario, a single physical machine’s resources can be shared by multiple virtual machines. How to map the physical resources to virtual machine resources and optimize the virtual resource allocation and scheduling has become a problem to be solved in the management of cloud computing resources.The main work and achievements:1. Establish three mathematical models for cloud computing services:variable-size service model, dynamic fixed-size service model, static fixed-size service model. Variable-size service model is the most similar to the real scene. In order to simplify the research process, we also proposes a fixed-size services model. In this model there is only one kind of physical machines. And according to the virtual machines’ time of leaving the system, there are dynamic fixed-size service model and static fixed-size service model.2. On the basis of the above three models,three scheduling algorithms based on combinatorial optimization are presented. The MS-V algorithm is presented for the variable-size service model in which there are more than one type of physical machines. Different costs are attached to different kinds of physical machines, the target is to minimize the cost of physical machine in the cluster. the MS-SF and MS-DF algorithm is proposed for the fixed-size service model. Since there is only one type of physical machine,the weights are the same and the target becomes the fewest number of physical machines.3. MS-SF is the basis of the rest two algorithms, MS-DF and MS-V are extensions of MS-SF. The enhancing configuration pair is introduced to enhance the optimization target. In chapter III of this article, it is demonstrated when the cluster state to deviate from the optimal solution in some distance, there must be an enhancing configuration pair to optimize the target.4. A prototype of virtual resources scheduling system is proposed based on the MS-series algorithms. The prototype has a master-slave structure. There is a controller node and multiple computing nodes.5. At the end of this paper, a simulation of the MS series is implemented using cloudsim. A comparison between the utilization of resources of MS series algorithms and classic heuristic algorithms. When the number of requests increases, MS series algorithms gain a higher utilization of resources.