Collaborative Research: Frameworks: Production quality Ecosystem for Programming and Executing eXtreme-scale Applications (EPEXA)

合作研究：框架：用于编程和执行超大规模应用程序的生产质量生态系统 (EPEXA)

• 批准号: 1931387
• 负责人: Robert Harrison
• 金额: $ 109.93万
• 依托单位: SUNY at Stony Brook
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2019
• 资助国家: 美国
• 起止时间: 2019-11-01 至 2024-10-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1931387&HistoricalAwards=false
• 关键词: Collaborative; Research; Frameworks; Production; quality

A team of researchers from three institutions will work collaboratively to design and develop a software framework that implements high-performance methods for irregular and dynamic computations that are poorly supported by current programming paradigms. The framework, titled EPEXA (Ecosystem for Programming and Executing eXtreme Applications), will create a production-quality, general-purpose, community-supported, open-source software ecosystem that attacks the twin challenges of programmer productivity and portable performance for advanced scientific applications on modern high-performance computers. Employing science-driven co-design, the team will transition into production a successful research prototype of a new programming model and accelerate the growth of the community of computer scientists and domain scientists employing these tools for their research. The project bridges the so-called "valley of death" between successful proofs of principle to an implementation with enough quality, performance, and community support to motivate application scientists and other researchers to adopt the tools and invest their own effort into the community. In addition to work on the framework development, the project includes training of postdoctoral scholars, graduate and undergraduate students as well as education, outreach and scientific community engagement activities.Specifically, the new powerful data-flow programming model and associated parallel runtime directly address multiple challenges faced by scientists as they attempt to employ rapidly changing computer technologies including current massively-parallel, hybrid, and many-core systems. Both data-intensive and compute-intensive applications are enabled in part by the general programming model and through the ability to target multiple backends or runtime systems. Also enabled is the creation by domain scientists of new domain-specific languages (DSLs) for both shared and distributed-memory computers. EPEXA contributes to the design and development of state-of-the-art software environments that leverage the National Science Foundation's investments in cyberinfrastructure to enable scientific discovery across all disciplines.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.

来自三个机构的研究人员团队将合作设计和开发一个软件框架，该框架可以实现当前编程范式很少支持的不规则和动态计算的高性能方法。该框架名为 EPEXA（编程和执行极限应用程序生态系统），将创建一个生产质量、通用、社区支持的开源软件生态系统，以应对高级科学应用程序的程序员生产力和可移植性能的双重挑战在现代高性能计算机上。 通过采用科学驱动的协同设计，该团队将把新编程模型的成功研究原型转变为生产，并加速使用这些工具进行研究的计算机科学家和领域科学家社区的发展。 该项目在成功的原理证明与具有足够质量、性能和社区支持的实施之间架起了所谓的“死亡之谷”，以激励应用科学家和其他研究人员采用这些工具并将自己的努力投入到社区中。除了框架开发工作外，该项目还包括对博士后学者、研究生和本科生的培训以及教育、推广和科学界参与活动。具体来说，新的强大的数据流编程模型和相关的并行运行时直接解决多个问题科学家在尝试采用快速变化的计算机技术（包括当前的大规模并行、混合和多核系统）时面临的挑战。数据密集型和计算密集型应用程序都部分地通过通用编程模型以及针对多个后端或运行时系统的能力来实现。领域科学家还可以为共享和分布式内存计算机创建新的领域特定语言（DSL）。 EPEXA 致力于设计和开发最先进的软件环境，利用美国国家科学基金会在网络基础设施方面的投资来实现跨所有学科的科学发现。该奖项反映了 NSF 的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准。

项目成果

Distributed-memory multi-GPU block-sparse tensor contraction for electronic structure

• DOI: 10.1109/ipdps49936.2021.00062
• 发表时间: 2020-06
• 期刊: 2021 IEEE International Parallel and Distributed Processing Symposium (IPDPS)
• 作者: T. Hérault;Y. Robert;G. Bosilca;R. Harrison;C. Lewis;Edward F. Valeev;J. Dongarra

Pushing the Boundaries of Small Tasks: Scalable Low-Overhead Data-Flow Programming in TTG

突破小任务的界限：TTG 中的可扩展低开销数据流编程

• DOI: 10.1109/cluster51413.2022.00026
• 发表时间: 2022
• 期刊: 2022 IEEE International Conference on Cluster Computing (CLUSTER
• 作者: Schuchart, Joseph;Nookala, Poornima;Herault, Thomas;Valeev, Edward F.;Bosilca, George
• 通讯作者: Bosilca, George

Understanding Recursive Divide-and-Conquer Dynamic Programs in Fork-Join and Data-Flow Execution Models

了解 Fork-Join 和数据流执行模型中的递归分治动态程序

• DOI: 10.1109/ipdpsw52791.2021.00069
• 发表时间: 2021
• 期刊: 2021 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW
• 作者: Nookala, Poornima;Ahmad, Zafar;Javanmard, Mohammad Mahdi;Kong, Martin;Chowdhury, Rezaul;Harrison, Robert
• 通讯作者: Harrison, Robert

Generalized Flow-Graph Programming Using Template Task-Graphs: Initial Implementation and Assessment

• DOI: 10.1109/ipdps53621.2022.00086
• 发表时间: 2022-05
• 期刊: 2022 IEEE International Parallel and Distributed Processing Symposium (IPDPS)
• 作者: J. Schuchart;Poornima Nookala;M. Javanmard;T. Hérault;Edward F. Valeev;G. Bosilca;R. Harrison

The Template Task Graph (TTG) - an emerging practical dataflow programming paradigm for scientific simulation at extreme scale

模板任务图（TTG）——一种新兴的实用数据流编程范例，用于极端规模的科学模拟

• 发表时间: 2020
• 期刊: 2020 IEEE/ACM 5th International Workshop on Extreme Scale Programming Models and Middleware (ESPM2
• 作者: Bosilca, G.;Harrison, R.J.;Herault, T.;Javanmard, M.M.;Nookala, P.;Valeev, E.F.
• 通讯作者: Valeev, E.F.