The Study And Implementation Of Information Service In MASSIVE Grid Environment

The Grid integrates geographically distributed, heterogeneous and dynamic resources, both hardware and software, into a uniform framework to provide low cost but high performance computational power. The grid information service, which is a vital part for any grid environment, provides a real-time image for grid resources and is a precondition for grid job schedule and job execution. In a grid, the distributed participators have autonomous controls of local resources and follow different policies to define their own sharing rules. Due to the distributed, large-scale, heterogeneous, dynamic and autonomous behaviors of resources, grid information service is a complex and challenging issue.The proposed system MEMDS (Extensible Monitoring and Discovery Service for the MASSIVE grid) is a part of MASSIVE (Multidisciplinary ApplicationS-oriented Simulation and Visualization Environment) project and studies information service for the MASSIVE Grid. The MASSIVE aims to establish a grid-enabled problem solving environment for distributed simulation and visualization of large-scale scientific and engineering research, such as Computational Fluid Dynamics (CFD) and Computational Solid Mechanism (CSM). The MASSIVE Grid is built on Globus Toolkit 2.X, and utilizes distributed resources to serve for mesh generation, distributed parallel computing, visualization and data analysis.The thesis focuses on the issues of grid information service and describes the design and implementation of MEMDS. MEMDS is a visualized and extensible resource discovery service and information service built on Globus MDS. The requirement and strategies of information service are discussed, and then, MEMDS is implemented based on GSI, SASL protocol and LDAP protocol to provide secure and convenient information services.In a grid, existing command-line tools are often inefficient and inconvenient for programmers. Therefore, MEMDS APIs are developed to simplify access to the MDS and provide a uniform interface by encapsulating LDAP operations. As a part of MASSIVE grid, these APIs will be called by higher-level services such as job monitoring service, and can be reuse in other programming implementation. Besides, a friendly graphic interface is developed based on these APIs to provide information query and monitoring.On the other hand, different applications require different resources, and thusinformation provided by default MDS service may be inadequate for specific engineering applications. To solve this problem and to support the MASSIVE grid, several custom Information Providers (IP) are implemented and integrated into MEMDS as a part of extention to MDS. To establish and dispatch custom information services easily, the MEMDS provides a service deployment module. This module consists of three parts: custom-IP, IP registry and deployment protocol, and the interface with the Globus MDS. Corresponding practices involved in the MASSIVE project show that the MEMDS improves the extensibility and flexibility of the grid information service.The thesis also discusses grid information service on PC clusters. As all slave nodes in a cluster have only local IP addresses, default information service can only gathers information about the master node of a cluster. Therefore, the information is illogical and a grid task may not be executed smoothly. This thesis proposes a solution by modifying and extending the Globus MDS to solve this problem and implements it.