SCI: ETF Early Operations: CACR/CIT

可扩展Terascale设备(ETF)或TeraGrid涉及九个资源提供商，它们运营高性能计算、信息和可视化设施，与专用光纤网络互连，并与一套通用软件集成，为用户社区提供一组连贯的资源和服务。2004年10月1日，TeraGrid将结束为期三年的建设工作，以创建这种分布式环境，并将过渡到运行状态。该提案概述了将由加州理工学院(CIT)的高级计算研究中心(CACR)作为TeraGrid的一部分提供的一系列服务。CACR将通过一种注重数据密集型应用的以科学为基础的方法来支持TeraGrid，最初来自共享以下特征的一小部分国家研究社区：o CIT有一个强大的相关科学研究小组；o其他机构有相关科学研究小组的领导人参与；o每个小组都反映了NSF或其他联邦研究机构的优先方向，主要是多机构感兴趣的领域。这些领域还将共享下列技术特征：o大量实验和/或计算数据保存在几个本地和远程不同的集合中；o需要新的分析方法来从实验数据中提取信息和/或与计算模型互动；o目前正在努力创建统一、可互操作的服务；o社区正在建立利用和联合信息资源的永久性虚拟组织。拟议的初始主要研究社区是天文学和高能物理。这两个领域都提供了与其他TeraGrid站点合作的机会，每个社区都有大量资金用于开发特定领域的网络基础设施。CACR将帮助这些小组在TeraGrid上部署和调整他们的软件，并提供支持以促进他们使用TeraGrid。此外，我们将与地球动力学建模和中子物理学社区一起开展工作。由于这些用户群体面临的问题有共同的特点，我们预计会确定共享网络基础设施的需求。CACR将运行一个由72个双处理器节点(0.8 Tf)组成的Itanium2集群。该群集通过Myrinet和千兆以太网与DataWulf群集紧密耦合，DataWulf群集将提供约170 TB的在线磁盘存储。这些资源将主要用于应用程序开发以及按需交互计算和数据分析。由于CACR有一个集中的角色，它可以响应最初的研究社区对按需和交互计算和数据分析的需求。固有的灵活性将允许对调度和一般资源管理的新方法进行试验。在过去的二十年里，CACR及其前身已经展示了与相对复杂的用户群体密切合作的能力，以探索支持计算科学前沿科学家的创新方法。我们建议继续这种合作模式。

The Extensible Terascale Facility (ETF), or TeraGrid, involves nine resource providers operating high-performance computing, information, and visualization facilities, interconnected with a dedicated optical network and integrated with a set of common software that presents the user community with a cohesive set of resources and services. On October 1, 2004, the TeraGrid will conclude a three-year construction effort to create this distributed environment and will transition into an operational state. This proposal outlines a set of services to be provided at the Center for Advanced Computing Research (CACR) of the California Institute of Technology (CIT) as part of the TeraGrid. CACR will support the TeraGrid through a science-based approach focused on data-intensive applications, initially from a small set of national research communities sharing the following features: o there is a strong related scientific research group at CIT; o there is participation by leaders of related scientific research groups at other institutions; and o each reflects a priority direction for NSF or other federal research agencies, primarily multi-agency areas of interest. These areas will also share the following technical attributes: o large quantities of experimental and/or computational data are maintained in several local and remote heterogeneous collections; o there are requirements for new analysis approaches to extract information from the experimental data and/or interact with computational modeling; o there is an ongoing effort to create unified, interoperable services; and o the communities are building persistent virtual organizations that exploit and federate information resources. The proposed initial primary research communities are astronomy and high energy physics. Both areas offer opportunities for collaboration with other TeraGrid sites and each community has significant funding to develop domain specific cyberinfrastructure. CACR will assist these groups in deploying and tuning their software on the TeraGrid and provide support to facilitate their TeraGrid usage. In addition, we will initiate efforts with the geodynamics modeling and neutron physics communities. Since the problems faced by these user communities have common features, we anticipate identifying requirements for shared cyberinfrastructure. CACR will operate an Itanium2 cluster consisting of 72 dual processor nodes (.8 TF). This cluster is tightly coupled via Myrinet and Gigabit Ethernet to DataWulf clusters, which will provide approximately 170 TB of online disk storage. These resources will be utilized primarily for application development and for on-demand and interactive computation and data analysis. Because CACR has a focused role, it can be responsive to the initial research communities requirements for on-demand and interactive computation and data analysis. Inherent flexibility will allow experimentation with novel approaches to scheduling and, in general, resource management. For the past two decades, CACR and its predecessors have demonstrated an ability to work closely with relatively sophisticated user groups to explore innovative approaches to supporting scientists on the frontier of computational science. We propose to continue this mode of collaboration.