Collaborative Research: Frameworks: Scalable Performance and Accuracy analysis for Distributed and Extreme-scale systems (SPADE)

协作研究：框架：分布式和超大规模系统的可扩展性能和准确性分析 (SPADE)

• 批准号: 2311707
• 负责人: Heike Jagode
• 金额: $ 210.27万
• 依托单位: University of Tennessee Knoxville
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2023
• 资助国家: 美国
• 起止时间: 2023-09-15 至 2027-08-31
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2311707&HistoricalAwards=false
• 关键词: Collaborative; Research; Frameworks; Scalable; Performance

Advances in computer simulations have made scientific discoveries more accessible. However, with the evolution of computing technology, each new generation of hardware and software presents unique performance and reliability challenges. These challenges must be addressed to fully harness the potential of these evolving technologies. SPADE is a project aimed to tackle these issues head-on. At its core, SPADE builds on the PAPI performance monitoring library - a tool used by the High-Performance Computing (HPC) community for over two decades. SPADE aims to enhance this legacy by creating methods that can assess and improve performance and accuracy on a wide range of advanced and evolving hardware and software technologies. This endeavor is not just about improving computational science but also about fostering diversity and education of a new generation of application scientists, engineers, and computer scientists. By providing an understanding of, and the ability to, navigate the intricate details of emerging technologies in the computing realm, SPADE is directly contributing to the advancement of this field. This will also democratize access to HPC, allowing a more diverse range of researchers and institutions to contribute to scientific discovery. Moreover, as SPADE aims to improve the capabilities of computer simulations, it enhances the ability to tackle a broad range of challenges - from understanding climate change to drug discovery. In essence, beyond advancing the HPC field, SPADE intends delivering a real-world impact by unlocking the full potential of computational science.The SPADE project focuses on advancing the monitoring, optimization, evaluation, and decision-making capabilities for extreme-scale systems. These critical capabilities are pivotal for both the High-Performance Computing (HPC) community and the scientific applications community that leverage these systems. With the evolution of HPC resources toward extreme scale, there is an increasing need for integrated performance and accuracy analysis frameworks to understand and mitigate performance and reliability challenges. To meet these needs, SPADE aims to deliver software and application programming interfaces (APIs) that broaden support for heterogeneity and scalability across a diverse range of computing platforms, including emerging vendor technologies. The SPADE project intends to utilize the established PAPI performance monitoring library to address the demands of scientific and machine learning applications effectively. Specifically, SPADE's mission includes: (1) developing monitoring capabilities for innovative and advanced technologies across the hardware stack; (2) designing novel abstractions that encapsulate the internal behavior of software components and facilitate interoperability across the software stack; (3) implementing a new performance and accuracy analysis framework that capitalizes on the efficiency and flexibility of C++'s object-oriented nature; (4) integrating new analysis functionality with various software stack layers and scientific and machine learning applications; and (5) examining new accuracy vs. performance trade-offs introduced with low-precision floating-point types. In essence, SPADE facilitates innovations in cyberinfrastructure development by enabling efficient and comprehensive resource utilization of extreme-scale platforms.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.

计算机模拟的进步使科学发现变得更容易获得。然而，随着计算技术的发展，每一代新的硬件和软件都提出了独特的性能和可靠性挑战。必须应对这些挑战，以充分利用这些不断发展的技术的潜力。Spade是一个旨在正面解决这些问题的项目。Spade的核心是建立在PAPI性能监控库的基础上--高性能计算(HPC)社区使用了二十多年的工具。Spade旨在通过创建方法来加强这一遗产，这些方法可以评估和提高各种先进和不断发展的硬件和软件技术的性能和准确性。这一努力不仅是为了改进计算科学，也是为了培养新一代应用科学家、工程师和计算机科学家的多样性和教育。通过提供对计算领域新兴技术的复杂细节的理解和导航能力，Spade正在直接为该领域的进步做出贡献。这也将使获得高性能计算的机会大众化，使范围更广的研究人员和机构能够为科学发现作出贡献。此外，由于Spade旨在提高计算机模拟的能力，它增强了应对从理解气候变化到药物发现等广泛挑战的能力。本质上，除了推进高性能计算领域，SPADE还打算通过释放计算科学的全部潜力来提供现实世界的影响。SPADE项目专注于推进极端规模系统的监控、优化、评估和决策能力。这些关键功能对于高性能计算(HPC)社区和利用这些系统的科学应用社区都至关重要。随着HPC资源向极大规模发展，越来越需要集成的性能和精度分析框架来了解和缓解性能和可靠性方面的挑战。为了满足这些需求，Spade的目标是提供软件和应用程序编程接口(API)，以扩大对各种计算平台(包括新兴供应商技术)的异构性和可扩展性的支持。SPADE项目旨在利用已建立的PAPI业绩监测库，有效地满足科学和机器学习应用程序的需求。具体地说，Spade的任务包括：(1)在整个硬件堆栈中开发对创新和先进技术的监控能力；(2)设计新颖的抽象，以封装软件组件的内部行为并促进软件堆栈之间的互操作性；(3)实现新的性能和精度分析框架，该框架利用C++的S面向对象的高效和灵活性；(4)将新的分析功能与各种软件堆栈以及科学和机器学习应用程序集成；以及(5)研究低精度浮点类型引入的新的精度与性能之间的权衡。本质上，SPADE通过实现对极端规模平台的高效和全面的资源利用，促进了网络基础设施开发的创新。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影响审查标准进行评估，被认为值得支持。