The Research On The Key Technologies Of Grid Information Service And Resource Management

Grid computing, which is considered to be one of the most important applications in next generation Internet, attracts much attention among researchers around the world. Grid computing is the strategic and national high performance computing and information infrastructure. The goal of grid computing is to integrate large-scale, decentralized and heterogeneous resources, such as super computers, storage systems, virtual reality systems, and instruments, through high speed networks into a virtual super computer to deal with large-scale application problems in scientific research, military areas, business application and information service. Now, the fact that grid computing is being merged with web service technology will enable grid computing the computation platform for business applications, which demand grid computing should seamlessly integrate current services and resources. Since the focus of grid computing is successfully moved from scientific research to business application, grid computing is capable of ubiquitous computing scheme, i.e. to compute anywhere and anytime.Grid security, information service, resource management and data management are the kernel technologies of grid computing. This dissertation focuses on the research on key technologies of grid interconnetion, information service and resource management, some of which are extended or improved by this dissertation. This dissertation makes key contributions to the following areas:1) An agent based grid interconnection method is proposed. In order to deal with security mechanism heterogeneity and inter-grid resource sharing problem, a grid interconnection method is proposed using Mobile Agent technology. This method is capable of inter-grid authentication, authorization, resource access control and other grid services. In this method some new ideas in grid resource load balance, fault tolerance, life-time management is illustrated. This method is built on layered hourglass grid architecture. A prototyped implementation is developed on Globus Toolkit 2 and Grasshoper. In order to evaluate the performance of this method, the theory of Stochastic Petri Net is used to model this method. Simulation results show this method is better than GSP method.2) A universal and fault-tolerant grid information services for multi-domain resource sharing is proposed. Grid computing depends on Grid Information Services (GIS) to support discovery and monitoring of the existence and characteristics of resources, services, computations and other entities. To date, GIS is VO (Virtual Organization) specific, which means GIS only provides services to VO members. The fact that VO members have little or no knowledge of resources outside of the VO they belongs to poses a significant obstacles to multi-domain resource sharing. By extending Globus MDS-2 (Monitoring and Discovery Service), a universal andfault-tolerant grid information services is proposed to discover, inquire and register multi-domain resources in a fault-tolerant way. For simplicity purpose, coarse-grained security mechanism is adopted on the level of VO to VO to tackle multi-domain security problem. In this architecture VO-specific aggregate directories join together to form a directory community and elect a Primary Directory to provide multi-domain grid information services to members of all VOs. Three key algorithms, Directory Community Discovery Algorithm, Primary Directory Election Algorithm and Community Partition Merging Algorithm are designed to link VOs together. The simulation results of the algorithms are also presented.3) A MDS-2 based, decentralized and robust wireless grid information service is proposed. On wireless grids not only users and service producers, but also aggregate directories are mobile and may not as powerful as those on wired grids, so information services are required to be decentralized and light-weighted. In this dissertation decentralized information service architecture, protocols, and algorithms are proposed to meet the special requirements on wireless grids. The proposed wireless grid information service extends the function of MDS-2 to support wireless grid resource sharing. This information service employs a group of aggregate directories on wireless grid. Grid information is dispatched to the group of directories, taking into account the distance between the directory and the resource provider, and processing power of the directories. To manage the group of directories and serve the users on wireless grids, a Primary Directory is selected among the group of directories. The architecture can tolerate failed or departed directories or resource providers, preventing them to crash the information service. To evaluate our approach, NS2 is used to simulate the three key algorithms, Directory Group Discovery Algorithm (DGDA), Primary Directory Election Algorithm (PDEA), Directory Dispatch Algorithm (DDA). The simulation results indicate that the proposed approach works well.4) A WSRF-compliant and P2P based information service on wireless grids is proposed. To further improve MDS-2 based, decentralized and robust wireless grid information service, meet the special requirements on wireless grids, immigrate to OGSA/WSRF architecture, and eliminate master-salve relationship, WGMDS, a WSRF-compliant and P2P (peer to peer) based information service is proposed. WGMDS extends the function of MDS-4 (Monitoring and Discovery Service) to support wireless Grid resource discovery, query and monitoring. WGMDS employs P2P mechanism to integrate information services to provide grid-wide resource information. In WGMDS, the clients and information sources dynamically probe nearest information services to query or register resource information. Three algorithms, Information Service Peers Discovery Algorithm, P2P Information Service Provision Algorithm, and Information Service Probing Algorithm are proposed to support the information service architecture. NS2 is used to evaluate the threealgorithms, and the experiment results show WGMDS works well. Some implementation issues are discussed as well.5) A virtual deadline based Min-Min scheduling heuristic for a meta-task is proposed. With the emergence of the computational grid, new scheduling algorithms are in demand to address concerns originating from the grid infrastructure, such as Heterogeneous Computing (HC) environment. HC environment is composed of a pool of machines with varied computational capabilities and a pool of applications with diverse computational requirements. Some heuristics are introduced to solve such scheduling problem, including Min-Min and other heuristics. A virtual deadline based Min-Min scheduling heuristic is propose to improve the performance of traditional Min-Min scheduling heuristic. The main idea of this heuristic is to introduce a virtual deadline into Min-Min scheduling heuristic. If the mapping of tasks to machines can be done under the deadline constraint, in fact, the virtual deadline can be removed from the mapping. If the virtual deadline is large enough, virtual deadline based Min-Min scheduling heuristic is identical to traditional Min-Min scheduling heuristic. In this heuristic, tasks are divided into 3 levels, T", T' and T. T" denotes those tasks which have one valid candidate machine under virtual deadline. T' denotes those machines the tasks which have at least one invalid candidate machine under virtual deadline. T denotes those machines which have not been mapped. The heuristic maps the tasks to machines in the order of T", T' and T. T" has the highest priority, T' higher priority, and T the lowest priority. In this heuristic, virtual deadline is set to be 1 or1.5 times of the makespan of traditional Min-Min scheduling heuristic. In this dissertation Rang based ETC(Expected Time to Compute) Generation method is used to generate ETC matrix. Combining machine heterogeneity and task heterogeneity, four sets of experiments are done to evaluate the proposed heuristic. The results show the proposed heuristic has less makespan than traditional Min-Min scheduling heuristic.