Next Generation Software: Supporting Complex Application Requirements in Metasystems

EIA-9974968 Grimshaw，Andrew S.弗吉尼亚大学下一代软件：支持元系统中的复杂应用程序需求拟议的工作解决了METASYSTEMS应用程序设计人员面临的两个关键问题：缺乏一套足够的工具来构建他们的应用程序、预测性能和做出调度决策；以及缺乏应用程序级容错技术。这项提议将通过建立我们过去使用Legion的经验来解决这些问题，Legion是一个用于设计和构建系统服务的体系结构，它向用户提供具有安全共享对象和共享名称空间的单个虚拟机的错觉。单个虚拟机可能由分布在多个管理域的数百个分布的、异构的资源组成，IDAEMA(集成设计、分析和执行元系统应用程序)是一个多阶段的建议和性能预测工具。在应用程序设计期间，IDAEMA使用基于应用程序特征和高级应用程序要求的编目研究成果库，以提供有关如何构建程序的分析和建议。该建议提供了多个视图，描述了设计选择对应用程序需求、软件工程成本和预测性能的影响--从如何开始在METASYSTEMS环境中编程的建议，到编码模板和现有组件，以此作为应用程序功能的基础。在运行时，IDAEMA使用有关目标元系统的硬件和软件的静态和动态信息，结合应用程序用户的安全性、容错性、可预测性和成本要求，在多个抽象和解析级别上提供帮助和性能预测。我们的容错方法是基于我们使用的RFE模型(反射图和事件)，并且确信没有一种单一的容错机制适合所有应用程序。相反，应用程序必须可以自由使用各种不同的技术，所有这些技术都在同一系统中。RGE提供了一种灵活的、可配置的机制，用于以程序图的形式将任意计算附加到事件。ExoEvents、ExoEvent兴趣和ExoEvent兴趣集用于动态配置特定于应用程序的容错技术。这两个组件共同为METASYSTEMS应用程序的开发和执行提供了新的灵活性和可预测性。IDAEMA和RGE模型通过对Legion的拟议修改与NPACI站点的合作者共同开发应用程序来验证，这些应用程序包括海洋-气候耦合模型和材料智能处理综合系统的全面模拟。

EIA-9974968Grimshaw, Andrew S.University of VirginiaNext Generation Software: Supporting Complex Application Requirements in Metasystems The proposed work addresses two critical problems facing metasystems applications designers: a lack of a sufficient set of tools by which to structure their applications, predict performance, and make scheduling decisions; and a lack of application-level fault-tolerance techniques. This proposal will address these problems by building upon our past experiences with Legion, which is an architecture for designing and building system services that provides to users the illusion of a single virtual machine with secure shared objects and shared name spaces. The single virtual machine may consist of hundreds of distributed and heterogeneous resources in multiple administration domains.IDAEMA (Integrated Design, Analysis, and Execution of Metasystems Applications) is a multi-stage advice and performance prediction tool for metasystems. During application design, IDAEMA uses a library of cataloged research results based on application characteristics and high-level application requirements to provide analysis and advice regarding how to structure the program. This advice provides multiple views that describe the implications of design choices on application requirements, software engineering costs, and predicted performance-from advice in how to get started programming in a metasystems environment to code templates and existing components upon which to base the functionality of the application. At run-time, IDAEMA uses static and dynamic information about the hardware and software of the target metasystem in combination with the application user's security, fault-tolerance, predictability and cost requirements to provide assistance and performance prediction at multiple levels of abstraction and resolution. Our approach to fault-tolerance is based upon our use of the RFE model (Reflective Graph and Event) and the conviction that no single fault-tolerance mechanism is appropriate for all applications. Instead application must be free to use a variety of different techniques, all in the same system. RGE provides a flexible, configurable mechanism for attaching arbitrary computation in the form of program graphs to events. Exoevents, Exoevent Interests, and Exoevent Interest Sets are used to dynamically configure application-specific fault-tolerance techniques. Together, these two components provide new flexibility and predictability for metasystems application development and execution. IDAEMA and the RGE model are validated through proposed modifications to Legion in combination with collaborators at NPACI site developing applications that include a coupled ocean-climate model and a full-scale simulation of an integrated system for intelligent processing of materials.