RI: Mastodon: A Large-Memory, High-Throughput Simulation Infrastructure

RI：Mastodon：大内存、高吞吐量的模拟基础设施

• 批准号: 0303609
• 负责人: Risto Miikkulainen
• 金额: $ 144.53万
• 依托单位: University of Texas at Austin
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2003
• 资助国家: 美国
• 起止时间: 2003-09-01 至 2008-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0303609&HistoricalAwards=false
• 关键词: RI; Mastodon; Large; Memory; Throughput

Digital simulation is becoming dominant in not only many fields of physical sciences, but computer science as well. Despite the great improvements in processing speed over the past decades, advanced research on large-scale systems requires simulation engines that are more powerful than the average desktop workstation.The Department of Computer Sciences at the University of Texas at Austin will assemble an infrastructure for performing large-scale, memory-intensive simulations. Called Mastodon, this infrastructure will consist of a large number of rack-mounted Linux/x86 servers, each with multiple processors and several gigabytes of physical memory. Leveraged by industrial and university matching funds, we anticipate scaling Mastodon to 418 processors and nearly a terabyte of DRAM. The cluster will utilize both distributed batch scheduling using the Condor job management software, as well as soft partitioning for timing and parallel experiments.This infrastructure will service the needs of a majority of our department's faculty. The projects utilizing Mastodon include the areas of systems (specifically, architecture, compilers, and run-time systems), network algorithms, computational biology, multi-agent robotics, and formal verification. In each of these areas, Mastodon makes multiple projects feasible, whereas with conventional computing infrastructure, most of these projects cannot even be attempted.Broader Impact: This unparalleled simulation resource will be made available to a broader community as well, including students attending summer camps to increase female enrollment in computer science, undergraduates for both educational class project and research use, and researchers in regional departments with insufficient resources to perform high-end simulations.

数字仿真不仅在物理科学的许多领域占据主导地位，而且在计算机科学领域也是如此。尽管在过去的几十年里，处理速度有了很大的提高，但对大规模系统的高级研究需要比普通桌面工作站更强大的仿真引擎。德克萨斯大学奥斯汀分校的计算机科学系将组装一个基础设施，用于执行大规模、内存密集型的模拟。这个基础架构被称为Mastodon，它将由大量安装在机架上的Linux/x86服务器组成，每个服务器都有多个处理器和几gb的物理内存。在工业和大学配套资金的推动下，我们预计将Mastodon扩展到418个处理器和近1tb的DRAM。集群将利用Condor作业管理软件的分布式批调度，以及用于定时和并行实验的软分区。这个基础设施将满足我们系大多数教员的需要。利用Mastodon的项目包括系统领域（特别是体系结构、编译器和运行时系统）、网络算法、计算生物学、多智能体机器人和形式化验证。在这些领域中，Mastodon使多个项目变得可行，而在传统的计算基础设施中，大多数项目甚至无法尝试。更广泛的影响：这种无与伦比的模拟资源也将提供给更广泛的社区，包括参加夏令营的学生，以增加计算机科学的女性入学率，用于教育课程项目和研究用途的本科生，以及资源不足的地区部门的研究人员进行高端模拟。