CRII: OAC: A Framework for Parallel Data-Intensive Computing on Emerging Architectures and Astroinformatics Applications

CRII：OAC：新兴架构和天文信息学应用的并行数据密集型计算框架

• 批准号: 1849559
• 负责人: Michael Gowanlock
• 金额: $ 17.5万
• 依托单位: Northern Arizona University
• 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: 2019-03-15 至 2022-02-28
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1849559&HistoricalAwards=false
• 关键词: CRII; OAC; Framework; Parallel; Data

The amount of data that needs to be analyzed by the scientific community is increasing due to growing volumes of data collected by current and future instruments and sensors. Consequently, scientific data analysis requires a significant amount of time. To decrease the amount of time needed to analyze data, methods need to utilize more computational resources, such as more processors in a computer. While the central processing unit (CPU) in a modern computer has traditionally been used to carry out data analysis, the past decade has seen an increase in using graphics processing units (GPUs) for data analysis. The modern graphics processing unit contains thousands of processors that can be used to execute a program faster than on the CPU. However, many algorithms for data analysis do not use the GPU to its full potential within the context of the broader computer system. This project advances a framework for understanding the performance of GPUs as applied to data analysis applications. The major goal of the project is to bridge the gap between algorithms that only use the GPU, and fully integrated algorithms that exploit the strengths of both the CPU and GPU. The ensemble of algorithms that are explored in the project support the needs of the astronomy community and researchers in other scientific areas that require efficient data analysis methods. The project aims to realize a new era in CPU/GPU computing that impacts both computer science and other scientific fields. An outcome of the project is the development of materials for educators teaching at the intersection of data analysis and parallel computing. This project includes mentoring K-12, undergraduate, and graduate students. Consequently, the project serves the national interest, as stated by NSF's mission, by promoting the progress of science, and to advance the national health and prosperity. New cyberinfrastructure, such as data analysis algorithms for emerging heterogeneous architectures, are needed to address cutting-edge scientific problems. Data analysis building blocks and algorithms have many data-dependent performance bottlenecks. New architectures have the potential to alleviate some of these key bottlenecks. However, the majority of GPU research minimally involves the CPU/host, and performs most of the computation on the GPU. This is a missed opportunity to more closely integrate both data and task parallelism between the CPU and GPU to simultaneously exploit concurrency across both architectures. This project examines a selection of key algorithms in the database, machine learning, data mining, and parallel computing communities. Using these algorithms, this project explores the continuum between GPU-only and mixed hybrid parallelism (data and task parallelism between the CPU and GPU) to identify key bottlenecks that can be reduced by exploiting underutilized resources. The selected algorithms are fundamental to scientific data processing workflows, and can advance time-domain astronomy cyberinfrastructure. The project integrates data-intensive computing insights into courses at the undergraduate and graduate levels, and pedagogical modules are developed to be used by instructors for teaching concepts of mixed (data and task) parallelism across the CPU and GPU. This project includes mentoring students at the undergraduate, graduate, and K-12 levels, including outreach at science festivals to encourage participation and interest in science, technology, engineering, and mathematics.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.

由于当前和未来的仪器和传感器收集的数据量不断增加，科学界需要分析的数据量正在增加。因此，科学数据分析需要大量时间。为了减少分析数据所需的时间，方法需要利用更多的计算资源，例如计算机中的更多处理器。虽然现代计算机中的中央处理单元 (CPU) 传统上用于进行数据分析，但在过去十年中，使用图形处理单元 (GPU) 进行数据分析的情况有所增加。现代图形处理单元包含数千个处理器，可用于比 CPU 更快地执行程序。然而，许多数据分析算法并未在更广泛的计算机系统环境中充分利用 GPU 的潜力。该项目提出了一个框架，用于了解应用于数据分析应用程序的 GPU 的性能。该项目的主要目标是弥合仅使用 GPU 的算法与充分利用 CPU 和 GPU 优势的完全集成算法之间的差距。该项目中探索的算法集合支持天文学界和其他科学领域研究人员需要高效数据分析方法的需求。该项目旨在实现 CPU/GPU 计算的新时代，影响计算机科学和其他科学领域。 该项目的成果是为教育工作者开发数据分析和并行计算交叉教学的材料。该项目包括指导 K-12、本科生和研究生。 因此，正如 NSF 的使命所言，该项目通过促进科学进步、促进国民健康和繁荣来服务于国家利益。需要新的网络基础设施，例如用于新兴异构架构的数据分析算法，来解决尖端科学问题。数据分析构建块和算法有许多与数据相关的性能瓶颈。新架构有潜力缓解其中一些关键瓶颈。然而，大多数 GPU 研究很少涉及 CPU/主机，并在 GPU 上执行大部分计算。这是一个错失的机会，可以更紧密地集成 CPU 和 GPU 之间的数据和任务并行性，从而同时利用两种架构之间的并发性。该项目研究了数据库、机器学习、数据挖掘和并行计算社区中的一系列关键算法。使用这些算法，该项目探索了纯 GPU 并行性和混合混合并行性（CPU 和 GPU 之间的数据和任务并行性）之间的连续性，以确定可以通过利用未充分利用的资源来减少的关键瓶颈。所选算法是科学数据处理工作流程的基础，并且可以推进时域天文学网络基础设施。 该项目将数据密集型计算见解集成到本科生和研究生级别的课程中，并开发了教学模块，供教师用于教授跨 CPU 和 GPU 的混合（数据和任务）并行性概念。该项目包括指导本科生、研究生和 K-12 级别的学生，包括在科学节上进行宣传，以鼓励对科学、技术、工程和数学的参与和兴趣。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进行评估，被认为值得支持。

项目成果

GPU-Accelerated Similarity Self-Join for Multi-Dimensional Data

• DOI: 10.1145/3329785.3329920
• 发表时间: 2018-09
• 期刊: Proceedings of the 15th International Workshop on Data Management on New Hardware
• 作者: M. Gowanlock;Ben Karsin

A study of work distribution and contention in database primitives on heterogeneous CPU/GPU architectures

异构 CPU/GPU 架构上数据库原语的工作分配和争用研究

• DOI: 10.1145/3412841.3441913
• 发表时间: 2021
• 期刊: SAC '21: Proceedings of the 36th Annual ACM Symposium on Applied Computing
• 作者: Gowanlock, Michael;Fink, Zane;Karsin, Ben;Wright, Jordan
• 通讯作者: Wright, Jordan

Heterogeneous CPU-GPU Epsilon Grid Joins: Static and Dynamic Work Partitioning Strategies

• DOI: 10.1007/s41019-020-00145-x
• 发表时间: 2020-10
• 期刊: Data Science and Engineering
• 影响因子: 4.2
• 作者: Benoît Gallet;M. Gowanlock

Data-Intensive Computing Modules for Teaching Parallel and Distributed Computing

用于并行和分布式计算教学的数据密集型计算模块

• DOI: 10.1109/ipdpsw52791.2021.00062
• 发表时间: 2021
• 期刊: 2021 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW
• 作者: Gowanlock, Michael;Gallet, Benoit
• 通讯作者: Gallet, Benoit

Hybrid KNN-join: Parallel nearest neighbor searches exploiting CPU and GPU architectural features

• DOI: 10.1016/j.jpdc.2020.11.004
• 发表时间: 2020-11
• 期刊: J. Parallel Distributed Comput.
• 作者: M. Gowanlock