Campus Grid Technology Research Satisfying QoS

As next generation Internet, Grid can integrate geographically distributed resource transparently, making itself especially fit for large-scaled distributed applications. Grid brings many advantages, including convenient and shortcut access interface, remote cooperation and resource sharing in virtual environment, and information island elimination. Research of Grid technology is valuable for education research, engineering and business applications.This thesis takes detail researches on grid system model and architecture, QoS mechanisms, resource management mechanisms, task scheduling algorithms, grid applications and grid testing, and achieves the following break through.(1) The campus grid specificity analysis: To simplify the studied object, we will first focus on the vision of colleges and universities' campus network. This paper has analyzed and studied all the special circumstances, which colleges and universities build a campus Grid, and presents that different algorithms can be adopted in the light-load and the heavy-load status.(2) QoS mechanisms' research for grid: According to the difference of QoS characteristics, which are expressed by QoS parameters, in virtual organizations layer Grid QoS parameters are classified, and the QoS measurement and realization methods are given, and finally the campus grid system framework are given. At the same time, aiming at the grid environment to ensure the quality of services is difficult, a set of practical resources, tasks and user classification mechanism are provided. The resources and tasks should be divided into high I/O required category, high-performance computing category and large memory category, and so on. When the resources firstly are added to Grid, it must provide performance parameter for Grid system. After the system verifies, the system will class and evaluate it. The task will be allocated to the resources, which is adept in completing this type of tasks. In executing process, the system renew to evaluate those resource based on the status of the task executing. On the other hand, Grid users will be divided into different priority based on the priority value levels, and the tasks will be serviced according to priority. When the priority is same, task's executing will use the method of first-come-first-service. Under this mechanism, the refusal rate is lower, but the satisfaction rate is high. On the basis of the above-mentioned, for improving the task scheduling methods, a dynamic region's grid resource management algorithms based on association rule are put forward. So, The QoS mechanisms of Grid are realized in a certain extent.(3) The research of grid services dependency: Grid service dependency analysis is a useful technique to solve problems such as grid service testing, maintenance, composition and failure detection in grid service mining (GSM). An approach based on graph theory and matrix model is proposed to analyze grid service dependencies. Possible types of grid service dependencies are investigated. Definitions of direct dependency graph (DDG), direct dependency matrix (DDM), dependency relation matrix (DRM), degree of direct dependency (DDD) and minimal cover (MC) are given out.(4) Grid task scheduling algorithm research: A heavy load situations in a dynamic rules-based correlation Fenwick resource management mechanisms, and on this basis were given a level dispatch center, dispatch center and secondary activation of the algorithm nodes mandate basic ideas and specific algorithms.(5) The research of the task syncopating and grouping model: In the paper, the sub-task grouping model based on association rule (SGMAR) was put forward. Firstly, the model began to analyze from the sub-task groups' support matrix and the confidence matrix, then worked out the Density Index value (DI) and the Coupling Index value (CI). By virtue of this pair index values, the model could minimize the coupling degree of the sub-task group finally. During the actual utilization, this model could optimize the sub-task grouping, but also forecasted the sub-tasks' available resources in the task performing. Consequently, the model achieved both to optimize the sub-task grouping and to increase the hit rate of the resources scheduling.(6) The campus grid building research using middleware: Discussed data exchange middleware based on Java technology, gave a specific architecture for campus electronic governmental affairs, presented how to realize the cross-platform data exchanging by XML data files, and so on.(7) Campus grid applications research: To solve the problem of software integration in campus grid application, a method that can realize the data's conversion and matching by java middleware based on XML was presented. Firstly, the data would be converted into XML document format by java middleware in old heterogeneous database. Then, the J2EE framework of Mule ESB based on open source software was given, and campus grid application was realized by organizing and setting up the integrated software system of SOA. Lastly, an application example was completed to evaluate the performance of the framework 's key technologies. The result shows the efficiency of the integration system.(8) The grid test research based on an extended petri net of EEPC: In the Grid testing, the integration testing was one of the most important. For cutting down the testing costs, the extended Petri net and EEPC model were taken as the researching base during the integration testing of components. After fusing adequately the excellences of Petri and EEPC in structure expression, qualitative and quantitative analysis etc, the PRI of components would be sorted through quantitatively analyzing some parameter, such as the using frequency of components etc. In the end, the testing emphases would be located in the components, which had the more upper PRI. Of course, the reason, why this method was brought forward, was in order to reduce the costs of testing in an across-the-board status.