SHF: Medium: Collaborative Research: Run-Time Support for Scalable Concurrent Programming

SHF：中：协作研究：可扩展并发编程的运行时支持

• 批准号: 1561807
• 负责人: Maurice Herlihy
• 金额: $ 54万
• 依托单位: Brown University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2016
• 资助国家: 美国
• 起止时间: 2016-09-01 至 2019-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1561807&HistoricalAwards=false
• 关键词: SHF; Medium; Collaborative; Research; Run

Highly-concurrent data structures lie at the heart of modern multicore software. This project will provide systematic run-time system support for several techniques that that have proved essential for constructing highly-concurrent data structures. The intellectual merits are to provide a new basis for thinking about how to scale future generations of both hardware and software, and in particular to develop novel uses of operating system kernel functionality, as well as transactional hardware and software techniques. The project's broader significance and importance is the benefit to society provided by higher performing, less expensive, and more reliable software.The specific techniques addressed are unsynchronized traversals, in which a thread navigates through a linked data structure without writing to memory, and atomic sequences of memory operations, where race conditions are eliminated by making a sequence of individual memory operations appear to take place instantaneously. Although these techniques have been successfully deployed in many ad-hoc instances, they have never been packaged as general-purpose mechanisms because they can have complex and dangerous interactions with standard memory management schemes. This project will exploit recent developments in hardware architectures and operating system structures to develop automated, systematic run-time support, making these techniques accessible to non-specialists.

高并发数据结构是现代多核软件的核心。该项目将为几种已被证明对于构建高并发数据结构至关重要的技术提供系统的运行时系统支持。其智力优点是为思考如何扩展未来几代硬件和软件提供了新的基础，特别是开发操作系统内核功能以及事务性硬件和软件技术的新颖用途。该项目更广泛的意义和重要性在于更高性能、更便宜和更可靠的软件为社会带来的好处。所涉及的具体技术是非同步遍历，其中线程在链接的数据结构中导航而不写入内存，以及内存操作的原子序列，其中通过使一系列单独的内存操作看起来瞬时发生来消除竞争条件。尽管这些技术已成功部署在许多临时实例中，但它们从未被打包为通用机制，因为它们可能与标准内存管理方案进行复杂且危险的交互。该项目将利用硬件架构和操作系统结构的最新发展来开发自动化、系统化的运行时支持，使非专业人士也可以使用这些技术。