Research On Service-oriented Grid Resource Management And Robustness

Because the grid has the features of distribution,heterogeneity and dynamic, the theory,mechanism and method of resource organization and manegement under the grid computingenviorment is the most kernel and challenging problem. Based on the comprehensivediscussion and intensive analysis of the the features of the grid resource management and theexisting system models, this paper studied and explored some key problems of the grid resourceallocation and robustness. The main contributions of the dissertation as follows:1) A service-oriented grid resource management model is built. The emergence of OGSArepresents a consensus on the standardization of resource accessing interfaces as grid services.However, current job management systems do not meet the application requirements from aservice grid. Therefore, a service-oriented grid resource management model is presented thatprovides a set of transparent and location independent interfaces. Service level is alsointroduced to differentiate QoS requirements. The paper proposes a service-oriented gridscheduling framework and gives a dynamic algorithm about job matching and scheduling. Thissolves the dynamic and resource competition problem in grid computing efficiently.2) Define computation method of the grid resource management. In order to minimize thecost of grid service while still maintaining an adequate quality of service we divide the gridservice cost into three parts: service deployment cost, service active cost and service wait cost.In this work we formulate the computation method of the costs. We use the dynamicprogramming method of value iteration for its solution and we characterize the resulting optimalcost function. It shows that the cost of providing a service is more sensitive to under-deployment than to over-deployment. Thus, when demand for the service is known to exist, butis unpredictable, it is better to allocate more resources than fewer resources.3) The paper introduces a model for deployment and hosting of a hierarchical gird servicewherein the service provider must pay a resource provider for the use of resources. The modelproduces policies that balance the number of required resources with the desire to keep the costof hosting the service to a minimum. We quantify the effect that users demand disturbancemakes the cost of reserved resources increase in each layer of our framework. A stochasticcontrol algorithm is cast in order to resolve the problem. 4) The robustness analysis and measurement method of grid resource allocation andscheduling are proposed in this dissertaion. Because of the loosely-coupling among the gridapplication components, the disturbance must lead system to instability. Therefore, we arguethat, in addition to performance metrics, the resource allocation algorithms should be designedfor robustness. That is, they should have the ability to maintain performance under a widevariety of operating conditions. We introduce the concept of robustness and present an analysistechnique to determine the robustness of grid resource allocation. As the main contributions ofthe paper, we present a mathematical description of a metric for the robustness of a resourceallocation with respect to a specific disturbance.