CISE Research Resources: A National Logistical Networking Testbed

EIA 02-24441 Beck，Micah D.Donarra，Jack J.；Livny，Mron；Plank，James S.；Wolski，田纳西大学诺克斯维尔分校的RichardCise RR：国家物流网络测试这个项目研究了基于互联网背板协议(IBP)和外部节点(ExNode)数据结构的分布式网络存储体系结构，探索了一种通信基础设施的方法，称为物流网络。该方法将社区资源共享的互联网模型应用于网络存储，为网络本身注入可共享、可扩展并可供外部调度的存储资源，就像IP数据报交付一样。统一数据传输和数据存储结果调度的共同管理的资源结构，为分布式应用程序之间的互操作状态管理创造条件。拟议的国家后勤网络试验床(NLNT)的资源吸引了在现实世界条件下对这项技术进行关键研究和测试所需的多种和大量使用。NLNT是一种实验性的通信结构，它在国家高性能网络上放置了大量可共享存储的“仓库”，供研究团体自由使用。这些仓库支持不同中间件和应用研究小组在有意义的负载下探索和测试物流网络技术。它的主要特征如下。NLNT1。存储：可扩展共享。为了帮助最大限度地共享有限时间的轻量级网络存储资源，NLNT使用了一种网络存储技术，即互联网背板协议(IBP)。IBP紧密模拟IP数据报服务，遵循可扩展共享的互联网设计原则，创建可在全球范围内扩展的共享存储服务。目标容量：1拍字节。计划中的项目资金和工业合作伙伴的贡献在第三年支持了100TB的NLNT能力。随着技术显示出价值，产能也会增长。部署：目标是无处不在。为了提供高性能的接入并支持大量使用，NLNT在两个主要的Gigapop上设置了大量的“仓库”，直接连接到国家主干。配置良好的本地仓库的高价值鼓励无处不在的存储资源调配。完全实现的NLNT的增长将由三个存储仓库提供：由Yotta Yotta捐赠的先进的11.68TB存储区域网络系统；一个专门用于高性能覆盖网络中间件研究的仅限RAM的专用存储仓库；以及第三个计划为88TB系统的存储集群，通过快速、无阻塞的交换机连接到网络。这些单元将部署在芝加哥的Gigapop(Starlight)和华盛顿特区附近(Max)。NLNT软件基金会IBP已经实现并进行了测试。基本软件组件(用于聚合分布式存储分配的xNode工具集、用于监视和预测分布式资源的性能的网络天气服务(NWS)、以及用于NLNT资源发现的后勤骨干(L骨))支持NLNT中间件研究，包括利用动态网络拓扑进行调度、为关键存储功能(例如，复制、缓存等)提供运行时支持的工具和方法、以及高性能覆盖网络技术。这项研究涉及加州大学圣巴巴拉分校(沃尔斯基分校)和威斯康星-麦迪逊分校(利夫尼分校)的合作者。

EIA 02-24441Beck, Micah D.Dongarra, Jack J.; Livny, Miron; Plank, James S.; Wolski, RichardUniversity of Tennessee - Knoxville CISE RR: A National Logistical Networking Testbed This project, investigating a distributed network storage architecture based on the Internet Backplane Protocol (IBP) and the external node (exNode) data structure, explores an approach to communication infrastructure, called Logistical Networking. This approach applies the Internet model of community resource sharing to network storage to infuse the network itself with storage resources that can be shared, scaled up, and exposed for external scheduling just like IP datagram delivery. A resource fabric that unifies the co-management of scheduling of data transport and data storage results, creating conditions for interoperable state management among distributed applications. The resources of the National Logistical Networking Testbed (NLNT) proposed attract the kind of diverse and heavy usage necessary to carry out critical research and testing on this technology under real world conditions. NLNT, an experimental communication fabric, places large "depots" of sharable storage on the nation's high performance network for the unbrokered use of the research community. These depots support the exploration and testing of Logistical Networking technology under meaningful loads by diverse middleware and application research groups. Its leading characteristics follow. NLNT1. Storage: scalably sharable. To help maximize the share use of time-limited lightweight allocations of network storage resources, NLNT uses a network storage technology, Internet Backplane Protocol (IBP). IBP closely models IP datagram service and follows Internet design principles for scalable sharing to create shared storage service that can scale globally.2. Target capacity: 1 Petabyte. Planned project funding and contributions from industrial partners support NLNT capacity of 100TB in year three. As the technology shows value, capacity grows.3. Deployment: aims for ubiquity. To provide high performance access and support heavy use, NLNT locates massive "depots" at two major Gigapops with direct access to the national backbone. High value of well-provisioned local depots encourages ubiquitous provisioning of storage.The growth of the fully realized NLNT will be seeded by three storage depots: an advanced 11.68TB Storage Area Network system donated by Yotta Yotta; a special purpose 16GB RAM-only storage depot to be used for middleware research on high performance overlay networks; and a third storage cluster planned as a 88TB system attached to the network via a fast, non-blocking switch. The units will be placed at Gigapops in Chicago (Starlight) and near Washington D.C. (MAX). NLNT software foundation, IBP, has already been implemented and tested. Basic software components (xNode toolset for aggregating distributed storage allocations, the Network Weather Service (NWS) for monitoring and forecasting the performance of distributed resources, and the Logistical Backbone (L-Bone) for NLNT resource discovery) support NLNT middleware research, including scheduling with dynamic network topology, tools and methodologies to provide runtime support for key storage functions (e.g., replication, caching, etc.), and high performance overlay network technology. The research involves collaborators from the Universities of California-Santa Barbara (Wolski) and Wisconsin-Madison (Livny).