SHF: Small: Dynamic Power Redistribution in Failure-Prone CMPs

SHF：小型：易发生故障的 CMP 中的动态功率重新分配

• 批准号: 0916821
• 负责人: David Albonesi
• 金额: --
• 依托单位: Cornell University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2009
• 资助国家: 美国
• 起止时间: 2009-09-01 至 2013-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0916821&HistoricalAwards=false
• 关键词: SHF; Small; Dynamic; Power; Redistribution

Future multi-core microprocessors will be capable of deconfiguring faulty units in order to permit continued operation in the presence of wear-out failures. However, the unforeseen downside is pipeline imbalance in which other portions of the pipeline are now overprovisioned with respect to the deconfigured functionality. Such an imbalance leads to sub-optimal chip-wide power provisioning, since power is now allocated to pipeline functions that no longer provide the benefit they did with a fully functioning chip.This research proposes to dynamically redistribute the chip power under pipeline imbalances that arise from deconfiguring faulty units. Through rebalancing -- achieved by temporary, symbiotic deconfiguration of additional functionality within the degraded core -- power is harnessed for use elsewhere on the chip. This additional power is dynamically transferred to portions of the multi-core chip that can realize a performance boost from turning on previously dormant microarchitectural features. The technical deliverables of this project will be: (1) a novel resilient multi-core system architecture -- including dynamic power redistribution management algorithms -- that achieves much higher performance than one that is oblivious to pipeline imbalances; and (2) detailed simulations that quantify this performance advantage for various multi-core workloads.The broader impacts of this project relate to integrated research and education, enhanced infrastructure for research, broad dissemination of results, and potential societal impact. Furthermore, the PI will recruit women and underrepresented minority students to work on the project.

未来的多核微处理器将能够解除故障单元的配置，以便在存在磨损故障的情况下继续操作。 但是，不可预见的下行是管道不平衡，在该管道的其他部分相对于解剖配置的功能过度提供了。这种不平衡会导致较优化的芯片范围的功率提供，因为现在将功率分配给管道功能，而这些功能不再为其功能齐全的芯片提供了好处。这项研究建议动态地重新分布芯片不平衡的芯片功率，而这些芯片功率是由削弱配置的错误单元所带来的。通过重新平衡 - 通过暂时的，共生的反构为降级核心内的其他功能 - 可以利用芯片其他地方使用的功率。这种额外的功率被动态地转移到多核芯片的一部分，这些芯片可以通过打开以前休眠的微体系特征来实现性能提升。 该项目的技术可交付成果将是：（1）一种新颖的弹性多核系统体系结构（包括动态功率重新分布管理算法），其性能要比忽略管道失衡的弹性要高得多。 （2）详细的模拟量化了各种多核工作负载的这种性能优势。该项目的更广泛影响与综合研究和教育有关，增强了研究基础架构的研究，广泛的结果传播以及潜在的社会影响。此外，PI将招募妇女和代表性不足的少数族裔学生从事该项目。