CAREER: D3: Addressing Emerging Data-Induced Challenges in Embedded and Real-Time Systems

职业：D3：解决嵌入式和实时系统中新出现的数据引发的挑战

• 批准号: 1750263
• 负责人: Cong Liu
• 金额: $ 53.84万
• 依托单位: University of Texas at Dallas
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2018
• 资助国家: 美国
• 起止时间: 2018-01-15 至 2022-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1750263&HistoricalAwards=false
• 关键词: CAREER; D3; Addressing; Emerging; Data

Data-driven embedded systems are here. The ability to create algorithms that process massive amounts of real-time sensor-captured data is enabling designers in many industries to develop intelligent embedded systems that automate actions and decisions, e.g., self-driving vehicles. This emerging data-intensive embedded computing paradigm brings a new set of data-induced challenges around guaranteeing timing predictability and enabling latency constraints to be analytically validated at design time. The goal of this research is to overcome challenges due to real-time processing of massive data in embedded systems in order to guarantee timing predictability. D3, a comprehensive resource management ecosystem with the capability of predictably processing massive real-time data-intensive workloads, is implemented in the operating system. D3 brings in a novel set of fundamental system-level techniques, which enable smart data filtering and characterization, transparent and supervised streaming for execution concurrency optimization, and predictable memory management under heterogeneous architectures. The hard algorithmic challenges due to co-scheduling memory and highly heterogeneous computing resources such that analytical guarantees on timing predictability become quantifiable will be addressed. Developing a comprehensive heterogeneous resource management ecosystem for predictably processing massive real-time data-intensive workloads would be a significant result for many application domains, such as transportation and robotics. The next-generation automotive system is a good example that could greatly benefit from the proposed research. The outcome of this project will pave the way to certifiability of safety-critical autonomous vehicles based on heterogeneous platforms. An innovative undergraduate research project to develop an educational tool that uses real-time data-driven system design concepts as its foundation assists in realizing the educational objectives. All source code and evaluation data will be freely available for direct download from public web servers maintained by the University of Texas at Dallas (UTD) Computer Science Department under an open source Gnu Public License. Additional case-study test programs and scripts will be freely available under the open source BSD license and directly downloadable from public web servers maintained by the UTD Computer Science Department. All data products produced in this project will also be archived in the UTD computer science computing repository for permanent storage. The repository is available at www.utdallas.edu/~cong/CareerRepo.

数据驱动的嵌入式系统就在这里。创建处理大量实时传感器捕获数据的算法的能力使许多行业的设计人员能够开发自动执行操作和决策的智能嵌入式系统，例如自动驾驶车辆。这种新兴的数据密集型嵌入式计算范式带来了一系列由数据引发的新挑战，围绕保证时序可预测性和在设计时对延迟约束进行分析验证。这项研究的目标是克服嵌入式系统中实时处理海量数据所带来的挑战，以保证时序可预测性。 D3是一个全面的资源管理生态系统，能够以可预测的方式处理大量实时数据密集型工作负载，并在操作系统中实现。 D3 引入了一套新颖的基本系统级技术，可实现智能数据过滤和表征、用于执行并发优化的透明和受监督的流式传输以及异构架构下的可预测内存管理。由于共同调度内存和高度异构计算资源而导致时序可预测性的分析保证变得可量化所带来的硬算法挑战将得到解决。开发一个全面的异构资源管理生态系统来可预测地处理大量实时数据密集型工作负载对于许多应用领域（例如交通和机器人）来说将是一个重要的结果。下一代汽车系统就是一个很好的例子，可以从拟议的研究中受益匪浅。该项目的成果将为基于异构平台的安全关键型自动驾驶汽车的可认证性铺平道路。一个创新的本科生研究项目，旨在开发一种以实时数据驱动的系统设计概念为基础的教育工具，有助于实现教育目标。所有源代码和评估数据都可以根据开源 Gnu 公共许可证从德克萨斯大学达拉斯分校 (UTD) 计算机科学系维护的公共网络服务器上免费直接下载。其他案例研究测试程序和脚本将在开源 BSD 许可证下免费提供，并可直接从 UTD 计算机科学系维护的公共网络服务器下载。该项目产生的所有数据产品也将归档在UTD计算机科学计算存储库中进行永久存储。该存储库位于 www.utdallas.edu/~cong/CareerRepo。

项目成果

NeuOS: A Latency-Predictable Multi-Dimensional Optimization Framework for DNN-driven Autonomous Systems

NeuOS：用于 DNN 驱动的自治系统的延迟可预测的多维优化框架

• DOI: 10.1007/978-3-642-41284-4_21
• 发表时间: 2024-09-14
• 期刊: Science China Information Sciences
• 作者: Soroush Bateni;Cong Liu
• 通讯作者: Cong Liu

Simulee: Detecting CUDA Synchronization Bugs via Memory-Access Modeling

Simulee：通过内存访问建模检测 CUDA 同步错误

• 发表时间: 2020-01
• 期刊: IEEE/ACM International Conference on Software Engineering
• 作者: Wu, Mingyuan;Ouyang, Yicheng;Zhou, Husheng;Zhang, Lingming;Liu, Cong;Zhang, Yuqun
• 通讯作者: Zhang, Yuqun

PIFA: An Intelligent Phase Identification and Frequency Adjustment Framework for Time-Sensitive Mobile Computing

PIFA：用于时间敏感移动计算的智能相位识别和频率调整框架

• DOI: 10.1109/rtas.2019.00013
• 发表时间: 2019-04
• 期刊: 2019 IEEE Real-Time and Embedded Technology and Applications Symposium (RTAS
• 作者: Zhang, Xia;Xiao, Xusheng;He, Liang;Ma, Yun;Huang, Yangyang;Liu, Xuanzhe;Xu, Wenyao;Liu, Cong
• 通讯作者: Liu, Cong

NeuOS: A Latency-Predictable Multi-Dimensional Optimization Framework for DNN-driven Autonomous Systems

NeuOS：用于 DNN 驱动的自治系统的延迟可预测的多维优化框架

• 发表时间: 2020-01
• 期刊: USENIX ATC
• 作者: Bateni, Soroush;Liu, Cong
• 通讯作者: Liu, Cong

Simulee: Detecting CUDA Synchronization Bugs via Memory-Access Modeling

Simulee：通过内存访问建模检测 CUDA 同步错误

• DOI: 10.1145/3377811.3380358
• 发表时间: 2020-06-27
• 期刊: 2020 IEEE/ACM 42nd International Conference on Software Engineering (ICSE)
• 作者: Mingyuan Wu;Yicheng Ouyang;Husheng Zhou;Lingming Zhang;Cong Liu;Yuqun Zhang
• 通讯作者: Yuqun Zhang