SHF: Small: Minimal Multithreading - Exploiting Redundancy in Parallel Systems

SHF：小：最小多线程 - 利用并行系统中的冗余

• 批准号: 1017578
• 负责人: Diana Franklin
• 金额: $ 49.92万
• 依托单位: University of California-Santa Barbara
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2010
• 资助国家: 美国
• 起止时间: 2010-08-01 至 2014-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1017578&HistoricalAwards=false
• 关键词: SHF; Small; Minimal; Multithreading; Exploiting

Due to technology and industry trends, parallelism and energyefficiency have emerged as critical considerations in future computingsystems. The investigators observe that many convenient models forwriting parallel programs can lead to significant redundancy incomputation and data. This redundancy can be used to improveperformance and reduce energy consumption.The investigators observe that many parallel software models(e.g. data parallelism, Single-Program Multiple Data (SPMD),multi-programming, and high-throughput computing) consist of multiplethreads of computation with very similar instructions streams andworking sets of data. The trick, however, is to design efficientmechanisms that both exploit this similarity and effectively supportthe differences. This project will follow two research thrusts toeliminate redundancy across five application domains. First, theresearch will explore multi-threaded processor core designs thateliminate redundant instruction fetch and/or execution. Second, theresearch will explore content-aware caching techniques that reduceredundant storage of identical data across parallel threads andprocesses.

由于技术和行业趋势，并行性和能效已成为未来计算系统的关键考虑因素。研究人员观察到，许多用于编写并行程序的方便模型可能会导致计算和数据的显著冗余。研究人员观察到，许多并行软件模型(如数据并行、单程序多数据(SPMD)、多程序设计和高吞吐量计算)由具有非常相似的指令流和工作数据集的多线程计算组成。然而，诀窍在于设计高效的机制，既利用这种相似性，又有效地支持差异。这个项目将遵循两个研究推力，以消除五个应用程序领域的冗余。首先，研究将探索多线程处理器核心设计，以消除冗余的指令获取和/或执行。其次，研究将探索内容感知缓存技术，以减少跨并行线程和进程的相同数据的冗余存储。