CRII: CSR: Enabling Efficient Real-Time Systems upon Multiple Parallel Resources

CRII：CSR：在多个并行资源上实现高效的实时系统

• 批准号: 1948457
• 负责人: Jing Li
• 金额: $ 17.5万
• 依托单位: New Jersey Institute of Technology
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2020
• 资助国家: 美国
• 起止时间: 2020-06-01 至 2024-05-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1948457&HistoricalAwards=false
• 关键词: CRII; CSR; Enabling; Efficient; Real

There is a technology trend to consolidate multiple applications onto a shared hardware platform to reduce the size, weight, power, and cost of real-time systems, such as self-driving vehicles and autonomous robots. Furthermore, modern platforms consist of Central Processing Units (CPUs) and Graphics Processing Units (GPUs) with an increasing number of processing cores that share resources. Moreover, many current applications, such as artificial intelligence applications, have high computation needs and must execute in parallel to satisfy their real-time constraints. These technology trends demand that real-time systems be able to schedule real-time applications upon the shared multiple parallel resources efficiently.This research will investigate new parallel real-time scheduling frameworks for modern platforms with multiple resources. The scheduling problem is classified into two categories: staged-resources scheduling for alternating usage of different types of resources (e.g., alternatively executing on CPUs and GPUs), and vectorized-resources scheduling for simultaneously using multiple types of resources (e.g., running on processing units that share the last-level cache). The project will establish new parallel real-time task models for the two categories of resource usages. Based on the models, novel real-time schedulers and their corresponding analyses will be developed to achieve the goal of efficient utilization of multiple resources. The project will advance the understanding of parallel scheduling in real-time systems and serves as the initial steps of the challenge of efficient parallel real-time systems upon powerful and complex modern platforms. This project can have industrial impact on a wide range of today's artificial intelligence-based real-time systems to improve their responsiveness, efficiency, and scalability. The project includes enriching outreach activities and diversity programs to promote Science, Technology, Engineering and Mathematics (STEM) educational activity and broaden participation in computing and engineering. Research products generated as part of this project will be retained, managed, and disseminated through resources available at the New Jersey Institute of Technology. The products will be preserved with the goal of storing them for at least three years after the completion of the project or the publication of the corresponding articles, whichever is later. The URL to the project repository is https://git.njit.edu/njit-prt.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.

目前的技术趋势是将多个应用程序整合到一个共享硬件平台上，以减少实时系统（例如自动驾驶车辆和自主机器人）的尺寸、重量、功耗和成本。此外，现代平台由中央处理单元 (CPU) 和图形处理单元 (GPU) 组成，具有越来越多的共享资源的处理核心。 此外，当前的许多应用程序，例如人工智能应用程序，具有很高的计算需求，必须并行执行才能满足其实时约束。这些技术趋势要求实时系统能够在共享的多个并行资源上高效地调度实时应用程序。本研究将研究用于具有多个资源的现代平台的新的并行实时调度框架。调度问题分为两类：交替使用不同类型资源的分阶段资源调度（例如，交替在CPU和GPU上执行），以及同时使用多种类型资源的矢量化资源调度（例如，在处理上运行）共享最后一级缓存的单元）。该项目将为两类资源使用建立新的并行实时任务模型。基于这些模型，将开发新颖的实时调度器及其相应的分析，以实现高效利用多种资源的目标。该项目将增进对实时系统中并行调度的理解，并作为在强大而复杂的现代平台上挑战高效并行实时系统的第一步。该项目可以对当今各种基于人工智能的实时系统产生工业影响，以提高其响应能力、效率和可扩展性。该项目包括丰富的外展活动和多样性计划，以促进科学、技术、工程和数学（STEM）教育活动并扩大对计算和工程的参与。 作为该项目一部分产生的研究产品将通过新泽西理工学院的可用资源进行保留、管理和传播。产品将被保存，目标是在项目完成或相应文章发表后（以较晚者为准）保存至少三年。项目存储库的 URL 为 https://git.njit.edu/njit-prt。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

Hard-Real-Time Routing in Probabilistic Graphs to Minimize Expected Delay

概率图中的硬实时路由可最大限度地减少预期延迟

• DOI: 10.1109/rtss49844.2020.00017
• 发表时间: 2020-12
• 期刊: RTSS
• 作者: Agrawal, Kunal;Baruah, Sanjoy;Guo, Zhishan;Li, Jing;Vaidhun, Sudharsan
• 通讯作者: Vaidhun, Sudharsan

A Survey of Machine Narrative Reading Comprehension Assessments

机器叙事阅读理解评估调查

• 发表时间: 2022-07
• 期刊: Survey Track of 31st International Joint Conference on Artificial Intelligence (IJCAI Survey Track
• 作者: Sang, Yisi;Mou, Xiangyang;Li, Jing;Stanton, Jeffrey;Yu, Mo
• 通讯作者: Yu, Mo

TVShowGuess: Character Comprehension in Stories as Speaker Guessing

TVShowGuess：故事中的人物理解作为说话者的猜测

• DOI: 10.48550/arxiv.2204.07721
• 发表时间: 2022-04-16
• 期刊: ArXiv
• 作者: Yisi Sang;Xiangyang Mou;Mo Yu;Shunyu Yao;Jing Li;Jeffrey Stanton
• 通讯作者: Jeffrey Stanton

RTGPU: Real-Time GPU Scheduling of Hard Deadline Parallel Tasks With Fine-Grain Utilization

RTGPU：利用细粒度利用硬期限并行任务的实时 GPU 调度

• DOI: 10.1109/tpds.2023.3235439
• 发表时间: 2023-05
• 期刊: IEEE Transactions on Parallel and Distributed Systems
• 影响因子: 5.3
• 作者: Zou, An;Li, Jing;Gill, Christopher D.;Zhang, Xuan
• 通讯作者: Zhang, Xuan

AMCilk: A Framework for Multiprogrammed Parallel Workloads

AMCIlk：多程序并行工作负载框架

• DOI: 10.1109/hipc50609.2020.00035
• 发表时间: 2020-12
• 期刊: HiPC
• 作者: Wang, Zhe;Xu, Chen;Agrawal, Kunal;Li, Jing
• 通讯作者: Li, Jing