CSR: Small: Collaborative Research: Reliability Driven Resource Management of Multi-Core Real-Time Embedded Systems

CSR：小型：协作研究：多核实时嵌入式系统的可靠性驱动资源管理

• 批准号: 1319784
• 负责人: Robert Dick
• 金额: $ 16.67万
• 依托单位: Regents of the University of Michigan - Ann Arbor
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2013
• 资助国家: 美国
• 起止时间: 2013-10-01 至 2018-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1319784&HistoricalAwards=false
• 关键词: CSR; Small; Collaborative; Research; Reliability

The project aims to improve the quality of service and lifetime of real-timeembedded systems, particularly those implemented on multi-core platforms. Incontrast with many other quality metrics, such as performance and powerconsumption, reliability is difficult to accurately estimate because it isinfluenced by design decisions, environmental conditions, and process variationduring integrated circuit fabrication.This project consists of three main technical tasks. (1) Develop a reliabilitymodeling and analysis framework that can efficiently and accurately determinethe impact of design and runtime management decisions on reliability. (2)Develop a reliability-driven resource management framework, which includesruntime algorithms for assignment and scheduling of real-time tasks to maximizesystem lifetime while keeping soft error rates low, and a lightweight techniqueto adaptively adjust the activation frequency of the algorithms (i.e.,overhead). (3) Develop wear state monitoring techniques and data collectioninfrastructure to enable the runtime refinement and validation of system-levelreliability models that require long-term in-field system deployment.The techniques developed in this project will support the production of morereliable and/or less expensive electronic devices, enabling integratedcircuits, which are susceptible to wear due to lifetime fault processes to beused in special-purpose computers with strict reliability and performancerequirements. In particular, this project aims to ease the use of multicoreprocessors in high-reliability computing applications with deadlines, such asautomotive, multimedia, and health care applications. These applications havehistorically seen slow adoption of multicore processors, despite their price,performance, and power consumption benefits. We believe this is partially dueto gaps in knowledge of how to control and optimize reliability on suchsystems, some of which this project will fill. The involvement of both industryresearchers and university students at the undergraduate and graduate levelwill result in a broad dissemination of the research results.

该项目旨在提高实时嵌入式系统的服务质量和生命周期，特别是那些在多核平台上实现的系统。与许多其他质量指标（如性能和功耗）相比，可靠性很难准确估计，因为它受到设计决策、环境条件和集成电路制造过程中工艺变化的影响。 (1)开发一个可靠性建模和分析框架，可以有效和准确地确定设计和运行时管理决策对可靠性的影响。 （2）开发一个可靠性驱动的资源管理框架，该框架包括用于分配和调度实时任务的运行时算法，以最大限度地延长系统寿命，同时保持低软错误率，以及自适应调整算法激活频率的轻量级技术（即，开销）。 (3)开发磨损状态监测技术和数据收集基础设施，以实现运行时的细化和验证需要长期现场系统部署的系统级可靠性模型。本项目开发的技术将支持生产更可靠和/或更便宜的电子设备，实现集成电路，其易于由于寿命故障过程而磨损，以用于具有严格可靠性和性能要求的专用计算机。 特别是，该项目旨在简化多核处理器在高可靠性计算应用中的使用，如汽车，多媒体和医疗保健应用。 尽管多核处理器在价格、性能和功耗方面具有优势，但这些应用程序对多核处理器的采用一直很缓慢。 我们认为这部分是由于在如何控制和优化此类系统的可靠性方面的知识空白，本项目将填补其中一些空白。工业研究人员和大学本科生和研究生的参与将导致研究成果的广泛传播。