Performance Debugging Support for Many-Core Processors

众核处理器的性能调试支持

• 批准号: 0903470
• 负责人: Milos Prvulovic
• 金额: $ 13.73万
• 依托单位: Georgia Tech Research Corporation
• 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: 2009-08-01 至 2012-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0903470&HistoricalAwards=false
• 关键词: Performance; Debugging; Support; Many; Core

Microprocessor performance has doubled approximately every 18 months for several decades, making existing applications faster without much programming effort and allowing more demanding new applications to be developed. However, now the trend of steadily improving performance for existing applications has stalled. The performance potential of a microprocessor chip is still rapidly increasing, but the increase now mainly comes from packing more and more processor cores onto the chip, not from making each core faster. This provides little benefit to existing applications that can only utilize one core, so they must be re-designed to take advantage of multiple cores. Unfortunately, correct design of multi-core applications is much harder to achieve than was the case for single-core applications, and it is even harder for programmers to achieve scalability ? enabling applications to keep improving their performance as even more cores become available in the future. This problem threatens to disrupt the entire hardware-software ecosystem and possibly put an end to rapid improvements in computing performance, the information technology revolution, and the resulting productivity increases.This research project will investigate hardware and software performance debugging mechanisms that would help programmers identify and alleviate performance problems in their many-core applications, both for use in application development and in education. This will help the information technology revolution stay on track by helping existing programmers and training new ones to write scalable many-core applications that will in turn create demand for even more cores, allowing rapid progress in computing performance to continue.

几十年来，微处理器性能大约每 18 个月翻一番，无需太多编程工作即可使现有应用程序运行速度更快，并允许开发要求更高的新应用程序。然而，现在稳定提高现有应用程序性能的趋势已经停滞。微处理器芯片的性能潜力仍在快速增长，但现在的增长主要来自于将越来越多的处理器内核封装到芯片上，而不是使每个内核变得更快。这对于只能利用一个核心的现有应用程序几乎没有什么好处，因此必须重新设计它们以利用多个核心。不幸的是，多核应用程序的正确设计比单核应用程序的情况更难实现，程序员实现可扩展性更难？随着未来更多内核的出现，应用程序能够不断提高其性能。这个问题有可能破坏整个硬件-软件生态系统，并可能终止计算性能的快速改进、信息技术革命以及由此产生的生产力提高。该研究项目将研究硬件和软件性能调试机制，帮助程序员识别并缓解其多核应用程序的性能问题，用于应用程序开发和教育。这将有助于信息技术革命保持正轨，帮助现有程序员并培训新程序员编写可扩展的多核应用程序，从而创造对更多内核的需求，从而使计算性能继续快速进步。