ITR/SY Collaborative Research: A Unified Relational Approach to Grid Information Services

点击翻译按钮获取中文摘要

An application running in a distributed computing environment such as the Computational Grid must adapt to the available hardware and software resources. This requires information about the properties of Grid resources such as hosts, network switches, links and paths, software libraries and systems, user and organization rights, software services, event channels and dictionaries, and more. The information needed for an application to run, the values of the information (how fast the information changes) and the freshness of the information (how fast updates must be pushed to the application) can vary dramatically. These attributes place significant demands on the resource information service, demands that are arising with increasing prevalence in the general area of directory services as well. The Grid Forum, an international standards body for world-wide Grid computing, is developing standards for representing and querying this information. There is much that is excellent about these evolving standards, but there are many forms of highly desirable queries that will be difficult or expensive to perform in these systems. In particular, dynamic information will require very high update rates not supported by LDAP-based implementations.This project will address these concerns through a proposed (and tentatively named) Grid Resource Information Service (GRIS), a unified relational approach to grid information services. The research will start with the full ACID (Atomicity/Consistency/Isolation/Durability) functionality of a relational database system and "build down" to a practical resource information system that still provides most of the benefits of the RDBMS. Such a system will provide a single highly flexible query model and language for all types of Grid resource information, no matter how dynamic. The research will culminate in an extensible implementation based on commodity database systems and the SQL language, including "canned queries" for non-SQL users. The project will evaluate the new system and techniques using logged updates and queries from an existing Grid information service, and comparing results with a hierarchical system such as Globus MDS2. To facilitate comparisons, the project will produce a set of benchmark queries from discussions with users, tool developers, and Grid Forum members, and will quantify the limits of these queries.