SHF: CSR: Small: A Cooperative Framework for Topology Awareness on Large-Scale Systems

SHF：CSR：小型：大型系统拓扑意识的合作框架

• 批准号: 1320125
• 负责人: Zhiling Lan
• 金额: $ 49.84万
• 依托单位: Illinois Institute of Technology
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2013
• 资助国家: 美国
• 起止时间: 2013-10-01 至 2017-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1320125&HistoricalAwards=false
• 关键词: SHF; CSR; Small; Cooperative; Framework

As the number of computer nodes increases, so does the size of the interconnect network. Historically, floating point was the most costly component of a system, but this is no longer the case. Systems today, and those anticipated in the future, are increasingly bound by their communication infrastructure and the power dissipation associated with data movement across the rapidly growing number of nodes. How to address the increasing cost of data movement on ever-growing systems is becoming critical. This project will develop a framework named COTA, a COoperative framework for Topology Awareness. COTA will be an integrated framework that coordinates across the hardware, job scheduler, runtime, and application to jointly attack the increasing concern of data movement for communication- and power-efficiency on large-scale systems. Most importantly, the framework will support topology awareness not only at job startup, but also during job execution. The newly developed mapping algorithms, topology-aware methods and tools, and topology-aware models will provide a critical foundation for the realization of topology awareness on current and future systems. This research will have a direct impact on system productivity as well as a broad range of application domains that use parallel systems for simulations. The project will also enhance the curriculum at Illinois Institute of technology, broaden the participation by underrepresented groups, and outreach to the surrounding communities.

随着计算机节点数量的增加，互连网络的规模也在增加。从历史上看，浮点数是系统中最昂贵的组件，但现在已经不是这样了。今天的系统和未来的系统越来越受到通信基础设施和与快速增长的节点之间的数据移动相关的功耗的限制。如何解决在不断增长的系统上不断增加的数据移动成本变得至关重要。该项目将开发一个名为COTA的框架，即拓扑感知合作框架。COTA将是一个集成的框架，协调硬件、作业调度器、运行时和应用程序，共同应对大规模系统中日益增长的通信和能效数据移动问题。最重要的是，该框架不仅在作业启动时支持拓扑感知，而且在作业执行期间也支持拓扑感知。新开发的映射算法、拓扑感知方法和工具以及拓扑感知模型将为当前和未来系统实现拓扑感知提供重要的基础。这项研究将对系统生产力以及使用并行系统进行模拟的广泛应用领域产生直接影响。该项目还将加强伊利诺伊理工学院的课程，扩大代表性不足群体的参与，并扩大到周边社区。