Computation for the Endless Frontier

无尽前沿的计算

• 批准号: 1818253
• 负责人: Daniel Stanzione
• 金额: $ 6000万
• 依托单位: University of Texas at Austin
• 依托单位国家: 美国
• 项目类别: Cooperative Agreement
• 财政年份: 2018
• 资助国家: 美国
• 起止时间: 2018-09-01 至 2025-02-28
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1818253&HistoricalAwards=false
• 关键词: Computation; Endless; Frontier

Computation is critical to our nation's progress in science and engineering. Whether through simulation of phenomena where experiments are costly or impossible, large scale data analysis to sift the enormous quantities of digital data scientific instruments can produce, or machine learning to find patterns and suggest hypothesis from this vast array of data, computation is the universal tool upon which nearly every field of science and engineering relies upon to hasten their advance. This project will deploy a powerful new system, called "Frontier", that builds upon a design philosophy and operations approach proven by the success of the Texas Advanced Computing Center (TACC) in delivering leading instruments for computational science. Frontier provides a system of unprecedented scale in the NSF cyberinfrastructure that will yield productive science on day one, while also preparing the research community for the shift to much more capable systems in the future. Frontier is a hybrid system of conventional Central Processing Units (CPU) and Graphics Processing Units (GPU), with performance capabilities that significantly exceeds prior leadership-class computing investments made by NSF. Importantly, the design of Frontier will support the seamless transition of current NSF leadership-class computing applications to the new system, as well as enable new large-scale data-intensive and machine learning workloads that are expected in the future. Following deployment, the project will operate the system in partnership with ten academic partners. In addition, the project will begin planning activities in collaboration with leading computational scientists and technologists from around the country, and will leverage strategic public-private partnerships to design a leadership-class computing facility with at least ten times more performance capabilities for Science and Engineering research, ensuring the economic competitiveness and prosperity for our nation at large.TACC, in partnerships with Dell EMC and Intel, will deploy Frontier, a hybrid system offering 39 PF (double precision) of Intel Xeon processors, complemented by 11 PF (single precision) of GPU cards for machine learning applications. In addition to 3x the per node memory of NSF's prior leadership-class computing system primary compute nodes, Frontier will have 2x the storage bandwidth in a storage hierarchy that includes 55PB of usable disk-based storage and 3PB of 'all flash' storage, to enable next generation data-intensive applications and support for the data science community. Frontier will be deployed in TACC's state-of-the-art datacenter which is configured to supply 30% of the system's power needs from renewable energy. Frontier will include support for science and engineering in virtually all disciplines through its software environment support for application containers, as well as through its partnership with ten academic institutions providing deep computational science expertise in support of users on the system. The project planning effort for a Phase 2 system with at least 10x performance improvement will incorporate a community-driven process that will include leading computational scientists and technologists from around the country and leverage strategic public-private partnerships. This process will ensure the design of a future NSF leadership-class computing facility that incorporates the most productive near-term technologies, and anticipates the most likely future technological capabilities for all of science and engineering requiring leadership-class computational and data-analytics capabilities. Furthermore, the project is expected to develop new expertise and techniques for leadership-class computing and data-driven applications that will benefit future users worldwide through publications, training, and consulting. The project will leverage the team's unique approach to education, outreach, and training activities to encourage, educate, and develop the next generation of leadership-class computational science researchers. The team includes leaders in campus bridging, minority-serving institute (MSI) outreach, and data technologies who will oversee efforts to use Frontier to increase the diversity of groups using leadership-class computing for traditional and data-driven applications.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.

计算对我们国家在科学和工程方面的进步至关重要。无论是通过模拟实验成本高昂或不可能的现象，还是通过大规模数据分析筛选科学仪器可以产生的大量数字数据，还是通过机器学习从大量数据中找到模式并提出假设，计算都是几乎每个科学和工程领域都依赖的通用工具，以加速其发展。该项目将部署一个强大的新系统，称为“前沿”，该系统建立在德克萨斯州高级计算中心（TACC）成功交付计算科学领先仪器的设计理念和操作方法的基础上。Frontier在NSF网络基础设施中提供了一个规模空前的系统，该系统将在第一天产生富有成效的科学，同时也为研究界在未来转向更强大的系统做好准备。 Frontier是传统中央处理器（CPU）和图形处理器（GPU）的混合系统，其性能大大超过了NSF之前的领先级计算投资。 重要的是，Frontier的设计将支持当前NSF领导级计算应用程序向新系统的无缝过渡，并支持未来预期的新的大规模数据密集型和机器学习工作负载。 在部署之后，该项目将与10个学术合作伙伴合作运营该系统。 此外，该项目将开始与来自全国各地的领先计算科学家和技术专家合作规划活动，并将利用战略性公私合作伙伴关系设计一个领先级的计算设施，其性能至少是科学和工程研究的十倍，确保我们国家的经济竞争力和繁荣。TACC与Dell EMC和Intel合作，将部署Frontier，这是一个混合系统，提供39 PF（双精度）的英特尔至强处理器，辅以11 PF（单精度）的GPU卡，用于机器学习应用。除了3倍于NSF之前领先的计算系统主要计算节点的每个节点内存外，Frontier还将在存储层次结构中拥有2倍的存储带宽，包括55 PB的可用磁盘存储和3PB的“全闪存”存储，以实现下一代数据密集型应用程序并支持数据科学社区。 Frontier将部署在TACC最先进的数据中心，该数据中心配置为可再生能源供应系统30%的电力需求。 Frontier将通过其对应用程序容器的软件环境支持，以及通过与10个学术机构的合作，为系统上的用户提供深入的计算科学专业知识，从而为几乎所有学科的科学和工程提供支持。第二阶段系统的项目规划工作至少有10倍的性能改进，将包括一个社区驱动的过程，将包括来自全国各地的领先的计算科学家和技术专家，并利用战略公私合作伙伴关系。 这一过程将确保未来NSF领导级计算设施的设计，该设施将采用最具生产力的近期技术，并为所有需要领导级计算和数据分析能力的科学和工程预测最有可能的未来技术能力。 此外，该项目预计将为领导级计算和数据驱动应用开发新的专业知识和技术，通过出版物、培训和咨询，使全球未来的用户受益。 该项目将利用该团队独特的教育、推广和培训活动方法，鼓励、教育和培养下一代领导级计算科学研究人员。该团队包括校园桥接、少数民族服务机构（MSI）外展和数据技术方面的领导者，他们将监督使用Frontier来增加使用领导级计算进行传统和数据驱动应用的群体的多样性的努力。该奖项反映了NSF的法定使命，并被认为值得通过使用基金会的智力价值和更广泛的影响审查标准进行评估来支持。

项目成果

Designing Hierarchical Multi-HCA Aware Allgather in MPI

• DOI: 10.1145/3547276.3548524
• 发表时间: 2022-08
• 期刊: Workshop Proceedings of the 51st International Conference on Parallel Processing
• 作者: Tu Tran;Benjamin Michalowicz;B. Ramesh;H. Subramoni;A. Shafi;D. Panda

Network-Assisted Noncontiguous Transfers for GPU-Aware MPI Libraries

GPU 感知 MPI 库的网络辅助非连续传输

• DOI: 10.1109/mm.2023.3241133
• 发表时间: 2023
• 期刊: IEEE Micro
• 影响因子: 3.6
• 作者: Suresh, Kaushik Kandadi;Khorassani, Kawthar Shafie;Chen, Chen Chun;Ramesh, Bharath;Abduljabbar, Mustafa;Shafi, Aamir;Subramoni, Hari;Panda, Dhabaleswar K.
• 通讯作者: Panda, Dhabaleswar K.

OMB-Py: Python Micro-Benchmarks for Evaluating Performance of MPI Libraries on HPC Systems

OMB-Py：用于评估 HPC 系统上 MPI 库性能的 Python 微基准

• DOI: 10.1109/ipdpsw55747.2022.00143
• 发表时间: 2022
• 期刊: 23rd Parallel and Distributed Scientific and Engineering Computing Workshop (PDSEC
• 作者: Alnaasan, Nawras;Jain, Arpan;Shafi, Aamir;Subramoni, Hari;Panda, Dhabaleswar K
• 通讯作者: Panda, Dhabaleswar K

Hy-Fi: Hybrid Five-Dimensional Parallel DNN Training on High-Performance GPU Clusters

Hy-Fi：高性能 GPU 集群上的混合五维并行 DNN 训练

• DOI: 10.1007/978-3-031-07312-0_6
• 发表时间: 2022
• 期刊: Proceedings International Conference on High Performance Computing
• 作者: Jain, A;Shafi, A.;Anthony, Q.;Kousha, P.;Subramoni, H.;Panda, DK.
• 通讯作者: Panda, DK.

Highly Efficient Alltoall and Alltoallv Communication Algorithms for GPU Systems

适用于 GPU 系统的高效 Alltoall 和 Alltoallv 通信算法

• DOI: 10.1109/ipdpsw55747.2022.00014
• 发表时间: 2022
• 期刊: Heterogeneity in Computing Workshop
• 作者: Chen, Chen-Chun;Khorassani, Kawthar Shafie;Anthony, Quentin G.;Shafi, Aamir;Subramoni, Hari;Panda, Dhabaleswar K.
• 通讯作者: Panda, Dhabaleswar K.