Design And Implementation Of Grid-based Computing Pool

The Research center of Electric Power System Automation Engineering (REPSAE) at UESTC is conducting the research to apply the grid computing technology for electric power system application, and developed PSGrid? (Power System Grid) based on Globus3.0. This platform manages the dispersed and heterogeneous computing resources with grid computing pool. Grid computing pool is a novel network model that performs centralized management of the grid resources. It connects the high-performance computers in different areas by high-speed network, integrates these resources by the special middleware, accepts computing requests from different areas through web interface, and schedules the tasks to the appropriate nodes. Hence, computing pool can improve the service quality of the resources and fully utilize them to satisfy the computing requirement.Because the resources in the grid vary dynamically, in order to make it possible for a user to select the appropriate nodes to execute the computing tasks according to the requests, monitoring the state of the resources in the computing pool is particularly important. Through monitoring, failure can be discovered timely, and eliminated. By analyzing the monitored data, the performance bottleneck can be found and the appropriate control measures to improve the system performance can be taken.In this thesis, the components of the computing pool are designed based on Globus platform. The service use LDAP (Lightweight Directory Access Protocol) to realize the effective management, monitoring and control of the resources in the computing pool. The monitored information includes the CPU utilization rate, the free memory and the network status of the current node, .etc. When a task is submitted to the computing pool through web interface, the system will run an effective algorithm to produce an optimal mapping scheme to ensure that the submitted tasks can be finished with minimal time, according to the synthetical performance index and the computing amount and the task amount obtained by task decomposition. The thesis mainly resolved the database security issue, the LDAP's accessing mechanism and the problem of how JNDI (Java Naming and Directory Interface) connects to the LDAP server, etc.The thesis tested the performance of the computing pool with the power system flow parallel computing problem, and compared the performance of the computing with and without the computing pool management. The results show that the computing pool can effectively manage the computing resources and improve the computing performance.