CRII: CSR: NeuroMC---Parallel Online Scheduling of Mixed-Criticality Real-Time Systems via Neural Networks

CRII：CSR：NeuroMC---通过神经网络实现混合关键实时系统的并行在线调度

• 批准号: 1755965
• 负责人: Zhishan Guo
• 金额: $ 17.49万
• 依托单位: Missouri University of Science and Technology
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2018
• 资助国家: 美国
• 起止时间: 2018-07-01 至 2018-10-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1755965&HistoricalAwards=false
• 关键词: CRII; CSR; NeuroMC; Parallel; Online

With progression of technology, as the transistors become smaller, more of them can be integrated into a semiconductor chip; this in turn enables integration of many processor "cores" into one chip. Emerging chips integrate different types of processor cores, which are specialized for various functions including graphics processing and pattern recognition. Availability of many cores creates a task scheduling problem for a software program, namely which portions of the software should execute on what type of core. Given that the capabilities of the cores are not equal, a task may take different amounts of time to finish depending on which core it was assigned to. In a real-time system, while mapping tasks to cores, deadlines should be met, which may not always be possible. Further, the consequences of missing a deadline are not the same for all tasks. Some are more forgiving than others. Consequently, tasks can be classified broadly based on their criticality. This research will investigate an efficient scheduler for mixed-criticality real-time systems. In a resource constrained system, consequences of failure to meet a deadline may range from catastrophic to Minor. Consequently, there are varying penalties associated with missing deadlines. The investigator plans to address the scheduling problem of mixed-criticality real-time systems. This scheduling problem is NP-hard; the proposed approach involves an artificial neural network (NN)-based scheduler. The investigator will experiment with parallel NNs for faster convergence, and develop a prototype system to evaluate the efficiency and scalability of this solution. The project is expected to lead to the ability to make near-optimal decisions in real time. This project serves as the initial step of a larger effort in bringing the parallel computation and neural networks together. Broad economic and societal impacts of this project include integration of research and education, as it will involve development of a new course offering and redesign of an existing course on real-time and cyber-physical systems. The project will involve undergraduate students in programming graphics processing units and will seek to support multiple female students to broaden participation. Research results, educational material, software, and experimental data will be disseminated on the project website.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.

随着技术的进步，晶体管变得越来越小，越来越多的晶体管可以集成到半导体芯片中；这反过来又使许多处理器“核心”集成到一个芯片中。新兴芯片集成了不同类型的处理器核心，专门用于各种功能，包括图形处理和模式识别。多核的可用性给软件程序带来了任务调度问题，即软件的哪些部分应该在什么类型的核上执行。考虑到核心的能力是不相等的，一个任务可能需要不同的时间来完成，这取决于它被分配到哪个核心。在实时系统中，当将任务映射到核心时，应该满足最后期限，这可能并不总是可能的。此外，错过最后期限的后果对所有任务来说都是不一样的。有些人比其他人更宽容。因此，任务可以根据其重要性进行广泛的分类。本研究将探讨混合临界实时系统的高效调度器。在资源受限的系统中，未能在截止日期前完成任务的后果可能从灾难性到轻微不等。因此，错过最后期限会受到不同的惩罚。研究者计划解决混合临界实时系统的调度问题。这个调度问题是np困难的；该方法涉及到一个基于人工神经网络（NN）的调度程序。研究人员将用并行神经网络进行实验，以加快收敛速度，并开发一个原型系统来评估该解决方案的效率和可扩展性。该项目有望实现实时做出近乎最佳决策的能力。这个项目是将并行计算和神经网络结合在一起的更大努力的第一步。该项目的广泛经济和社会影响包括研究和教育的整合，因为它将涉及开发新的课程，并重新设计现有的实时和网络物理系统课程。该项目将让本科生参与图形处理单元的编程，并将寻求支持多名女学生扩大参与。研究成果、教材、软件和实验数据将在项目网站上发布。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。