Collaborative Research: Automatic Extraction of Parallel I/O Benchmarks from HEC Applications

协作研究：从 HEC 应用程序中自动提取并行 I/O 基准

• 批准号: 0937571
• 负责人: Kai Shen
• 金额: $ 24.6万
• 依托单位: University of Rochester
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2009
• 资助国家: 美国
• 起止时间: 2009-09-15 至 2013-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0937571&HistoricalAwards=false
• 关键词: Collaborative; Research; Automatic; Extraction; Parallel

I/O performance is often an issue for high-end computing (HEC) codes, due to their increasingly data-intensive nature and the ever-growing CPU-I/O performance gap. Portable parallel I/O benchmarks can help (1) application developers to improve their codes' performance, (2) HEC storage systems architects to improve their designs, and (3) future and current owners of HEC platforms to reduce hardware cost and improve application performance through better system provisioning and configuration. To keep up with the growing scale and complexity of HEC applications, this project develops automated generation of parallel I/O benchmarks, analogous to the SPEC and NAS benchmarks for computation. Our approach will be embedded in BenchMaker, a prototype tool that takes a real-world, large-scale parallel application and automatically distills it into a compact, human-intelligible, I/O-intensive, and parameterized benchmark. Such a benchmark accurately reflects the original application's I/O characteristics and I/O performance, yet with shorter execution time, reduced need for libraries, better portability, and easy scalability. Benchmarks and tools that benefit the computational science community at large will be produced by this research. These benchmark prototypes will be used for parallel computing course projects and student research contests.

I/O性能通常是高端计算（HEC）代码的一个问题，因为它们的数据密集性越来越高，CPU-I/O性能差距也越来越大。可移植并行I/O基准测试可以帮助（1）应用程序开发人员提高代码性能，（2）HEC存储系统架构师改进设计，以及（3）HEC平台的未来和当前所有者通过更好的系统供应和配置来降低硬件成本并提高应用程序性能。为了跟上HEC应用程序不断增长的规模和复杂性，该项目开发了并行I/O基准测试的自动生成，类似于SPEC和NAS计算基准测试。我们的方法将嵌入BenchMaker中，这是一个原型工具，它采用真实世界的大规模并行应用程序，并自动将其提取为紧凑，易于理解，I/O密集和参数化的基准。这样的基准测试准确地反映了原始应用程序的I/O特性和I/O性能，但执行时间更短，对库的需求减少，可移植性更好，易于扩展。这项研究将产生有利于计算科学界的基准和工具。这些基准原型将用于并行计算课程项目和学生研究竞赛。