MRI Collaborative: Development of ESPRIT - Emerging systems' performance and energy evaluation instruments and testbench

未来的计算节点很可能依赖于异构处理和存储系统以及网络技术。为给定的应用程序确定最合适的计算系统需要在尽可能多的备选方案上评估应用程序的性能，这是一项繁琐的任务。该项目开发了ESPRIT（新兴系统性能和能源评估仪器和测试台），这是一个能够评估最适合特定应用类别的系统的计算系统。如果可以根据应用程序在广泛的性能特征上的相似性将它们分类，那么就有可能确定最适合特定应用程序类别的系统。这项工作将有助于大规模计算系统配置为更有效的操作和更低的能源使用。ESPRIT项目由最先进的计算节点组成；系统、内存、电源和能量模拟器；来自不同应用程序的基准；一套测量仪器；研究应用程序行为的模型；统计聚类和其他机器学习技术。这个项目的优点在于开发工具来评估应用程序的一些行为的性能特征，并将它们分类成集群，以便通过不同的能力和技术规模来确定最适合的能源效率设计。ESPRIT可以用来研究新的设计选择，或者为特定的设计调整应用程序。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Future computing nodes will most likely rely on heterogeneous processing and memory systems as well as networking technologies. Identifying the most suitable computing system for a given application requires the cumbersome task of evaluating the application's performance on as many alternatives as possible. This project develops ESPRIT (Emerging Systems PeRformance and Energy Evaluation Instrument and Testbench), a computing system capable of evaluating the most suitable system for specific classes of applications. If applications can be classified into groups based on their similarities along a wide range of performance characteristics, it may be possible to determine the system best suited for a specific class of applications. This work will help large-scale computing systems be configured for more efficient operation and lower energy use.The ESPRIT project consist of state of the art computing nodes; system, memory, and power and energy simulators; benchmarks from different applications; a suite of measuring instruments; models for investigating application behaviors; statistical clustering and other machine learning techniques.The merit of this project resides in the development of instruments to evaluate applications along a number of performance characteristics of behaviors and classifying them into clusters in order to identify the most suitable design for energy efficiencies by varying capacities as well as technology scales. ESPRIT could be used to investigate new design choices, or tune applications for specific designs.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.