MRI: Development of Apt, A Testbed Instrument with Adaptable Profiles for Network and Computational Science

提案#:1338155PI(s): Ricci, Robert；史蒂文·C·科尔巴托；埃里克·艾德；胡里奥·法切利；题目：MRI/Dev: Apt，一种具有网络和计算科学适应性配置文件的测试平台仪器项目建议：该项目，开发一种新型的适应性仪器（Apt），旨在实现一种新的控制框架，以管理计算节点集群、存储服务器和灵活的互连结构。适应性构成了为计算机科学研究创造各种仪器的关键特征。这项工作旨在代表新一代的测试平台工具，通过授权用户根据自己的需要进行定制，从而吸引更广泛的用户社区。该仪器有望利用和推动云计算的趋势，从而使可适应仪器环境的开发更容易、更便宜、风险更低。同时，它还提供了商业云所不具备的特性，包括科学仪器所需的资源保真度和透明度。该项目还开发了用于创建和使用配置文件的控制和管理系统，以及用于支持并发实验和以动态方式在它们之间分配资源的系统。Apt的基本构建模块包括：—资源配置、调度和调节；-运行状态跟踪；实验管理、存档、回放；和-用户管理。这项工作有助于使这些元素以一种可重用的方式实现，这种方式使得被提议的元素能够被支持广泛研究领域的概要文件共享。目前的目标领域包括，但不限于，可持续能源控制系统，生物分子科学，隐私和安全，以及数据密集型计算。更广泛的影响：该仪器有望成为计算机科学和其他基于计算的领域的国家资源，可通过互联网访问并为国家做出贡献。计算机基础。广泛和多样化的目标研究社区包括强大的经验和实验成分。该仪器非常适合于培训和教育。

Proposal #: 1338155PI(s): Ricci, Robert; Corbato, Steven C; Eide, Eric N; Facelli, Julio C; Van Der Merwe, JacobusInstitution: University of Utah Title: MRI/Dev: Apt, a Testbed Instrument with Adaptable Profiles for Network and Computational ScienceProject Proposed:This project, developing a novel adaptable instrument (Apt), aims at attaining a novel control framework to manage a cluster of compute nodes, storage servers, and a flexible interconnect fabric. Adaptability constitutes the key feature enabling the creation of diverse instruments for computer science research. The work is geared towards representing a new generation of testbed instruments that engage a broader community of users by empowering them to customize it to their needs. The instrument is expected to leverage and drive the trends in cloud computing, thus making the development of adaptable instrument environments easier, cheaper, and less risky. At the same time, it offers properties not available on commercial clouds, including the proposed levels of resource fidelity and transparency required of a scientific instrument. The project also develops the control and management system for creating and using profiles, and for supporting concurrent experiments and allocating resources among them in a dynamic manner. The fundamental building blocks of Apt identified include: - Resource provisioning, scheduling, and conditioning; - Operational state tracking; experiment management, archive, and replay; and - User management. This work contributes to make these elements reusable in a way that enables the proposed to be shared by profiles supporting a wide range of research domains. The domains currently targeted include, but are not limited to, sustainable energy control systems, biomolecular science, privacy and security, and data-intensive computation.Broader Impacts: The instrument is expected to be a national resource for computer science and other computing-based fields, Internet-accessible and to contribute to the nation?s cyberinfrastructure. The broad and diverse targeted research communities include strong empirical and experimental components. The proposed instrument is highly suitable for training and education.