OAC Core: Cost-Adaptive Monitoring and Real-Time Tuning at Function-Level

OAC核心：功能级成本自适应监控和实时调优

• 批准号: 2402542
• 负责人: Yinzhi Wang
• 金额: $ 42.65万
• 依托单位: University of Texas at Austin
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2024
• 资助国家: 美国
• 起止时间: 2024-08-01 至 2026-07-31
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2402542&HistoricalAwards=false
• 关键词: OAC; Core; Cost; Adaptive; Monitoring

This project aims to address the challenge of performance monitoring on supercomputers by developing a tool that provides function-level insights with minimal overhead, enabling real-time tuning of applications. The initiative addresses the gap in understanding computational practices within diverse scientific domains, thus aiding in informed decision-making for system design and numerical library optimization. This advancement promises to enhance the efficiency of existing supercomputing infrastructures and contributes to the NSF's mission by supporting scientific progress and educational diversity, ultimately catalyzing a broader spectrum of scientific breakthroughs.This project is designed to improve performance monitoring within high-performance computing. It aims to address the increasing complexity and diversity of applications spanning scientific research, engineering, big data, and artificial intelligence. The approach involves implementing function-level monitoring through dynamic binary instrumentation and managing the monitoring overhead with a heartbeat mechanism. Additionally, it integrates real-time tuning capabilities for optimizing numerical libraries at runtime. This endeavor seeks to enhance traditional job-level resource utilization monitoring tools significantly. The research will identify standard function calls, evaluate the instrumentation overhead, and develop and validate policies for controlling overhead and accuracy. It will also involve creating a performance benchmark for assessing real-time tuning. The intellectual merit of this project stems from its potential to provide a novel tool that offers a more precise resolution of application behaviors and enables real-time performance tuning. By introducing adaptive monitoring and real-time tuning at the function level for large computational platforms, this project aims to accelerate scientific progress. Furthermore, it promotes diversity and inclusivity by actively involving underrepresented minority groups, contributing to a more diverse and skilled workforce in high-performance computing.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.

该项目旨在通过开发一种工具来解决超级计算机性能监控的挑战，该工具能够以最小的开销提供功能级见解，从而实现应用程序的实时调整。该倡议解决了理解不同科学领域内计算实践的差距，从而有助于系统设计和数值库优化的明智决策。这一进步有望提高现有超级计算基础设施的效率，并通过支持科学进步和教育多样性，最终促进更广泛的科学突破，为 NSF 的使命做出贡献。该项目旨在改善高性能计算中的性能监控。它旨在解决科学研究、工程、大数据和人工智能领域日益复杂和多样化的应用。该方法涉及通过动态二进制检测实现功能级监控，并通过心跳机制管理监控开销。此外，它还集成了实时调整功能，可在运行时优化数值库。这项工作旨在显着增强传统的工作级资源利用率监控工具。该研究将确定标准函数调用，评估仪器开销，并开发和验证控制开销和准确性的策略。它还将涉及创建用于评估实时调整的性能基准。该项目的智力优势源于其提供一种新颖工具的潜力，该工具可以更精确地解决应用程序行为并实现实时性能调整。通过在大型计算平台的功能层面引入自适应监控和实时调整，该项目旨在加速科学进步。此外，它还通过积极让代表性不足的少数群体参与来促进多样性和包容性，为高性能计算领域培养更加多元化和熟练的劳动力队伍做出贡献。该奖项反映了 NSF 的法定使命，并通过使用基金会的智力优势和更广泛的评估进行评估，被认为值得支持。影响审查标准。