CI-ADDO-NEW: Manifold - A Scalable Simulation Infrastructure for Future Many Core Systems

CI-ADDO-NEW：Manifold - 适用于未来多核心系统的可扩展仿真基础设施

• 批准号: 0855110
• 负责人: Sudhakar Yalamanchili
• 金额: $ 129.84万
• 依托单位: Georgia Tech Research Corporation
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2009
• 资助国家: 美国
• 起止时间: 2009-09-01 至 2019-02-28
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0855110&HistoricalAwards=false
• 关键词: CI; ADDO; NEW; Manifold; Scalable

As the computing industry accelerates into the many core era, it is confronted with the need for reliable and accurate simulation, modeling and performance prediction tools for future systems comprised of tens to hundreds of cores per die. Due to different intellectual histories and goals, the integration of mature existing point tools into accurate models of even small 4-8 core systems has proven to be time consuming, error prone, and brittle, with the models themselves taking hours to days for a single simulation run on state-of-the art workstations. Concurrently, these effects make re-use across groups and institutions difficult and have begun to severely inhibit research and teaching progress in many core architectures. The proposed program of work will create an open source many core simulation and modeling infrastructure. This infrastructure, Manifold, will adapt techniques for concurrent simulation to enable existing highly parallel architectures to be harnessed to model future designs. Manifold will enable users to incorporate existing simulation tools when they exist, as well as provide models for key many core components and phenomena, for example interconnection networks. The goal is a software infrastructure that is i) easily accessible to all academic and research communities, ii) runs on commodity clusters, iii) runs on many core laptops for classroom use, and iv) reduces barriers to entry in catalyzing new efforts. Central to Manifolds mission is the development of curricular material for computer architecture modeling and simulation including new courses, course modules, training materials, and significant evangelization.

随着计算行业加速进入多核时代，每个芯片包含数十到数百个核心的未来系统需要可靠且准确的仿真、建模和性能预测工具。由于不同的知识历史和目标，将成熟的现有单点工具集成到即使是小型 4-8 核系统的精确模型中也被证明是耗时、容易出错且脆弱的，模型本身需要花费数小时到数天的时间才能完成单个任务。模拟在最先进的工作站上运行。同时，这些影响使得跨团体和机构的重用变得困难，并开始严重抑制许多核心架构的研究和教学进展。 拟议的工作计划将创建一个开源的多核心仿真和建模基础设施。 Manifold 这个基础设施将采用并发仿真技术，使现有的高度并行架构能够用于对未来设计进行建模。 Manifold 将使用户能够合并现有的仿真工具（如果存在），并为许多关键的核心组件和现象（例如互连网络）提供模型。目标是建立一个软件基础设施，i）所有学术和研究社区都可以轻松访问，ii）在商品集群上运行，iii）在许多核心笔记本电脑上运行以供课堂使用，iv）减少促进新努力的进入壁垒。 Manifolds 使命的核心是开发计算机架构建模和仿真的课程材料，包括新课程、课程模块、培训材料和重要的宣传。