CSR: Small: Fluid Communication for Parallel Programs

CSR：小：并行程序的流畅通信

• 批准号: 1117135
• 负责人: Martha Kim
• 金额: $ 20万
• 依托单位: Columbia University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2011
• 资助国家: 美国
• 起止时间: 2011-08-01 至 2013-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1117135&HistoricalAwards=false
• 关键词: CSR; Small; Fluid; Communication; Parallel

Every aspect of parallel software development is more complicated than forserial programs. This research focuses on one of the primary sources ofcomplexity: intra-application communication. Currently it is a programmer'sresponsibility to find an efficient mapping of their application'scommunication patterns onto the communication infrastructure of the targetsystem. This research flips that responsibility by developing a flexiblecommunication architecture and associated tools and algorithms that allow thetarget platform to be specialized for a particular application, rather thanvice versa. In addition to reducing the programmer's burden, specializationhas the potential to improve communication efficiency while the automatedtechniques can increase portability. This research poses questions whose answers have consequences at severallevels of the traditional system stack: Can programmers be freed fromhardware-specific optimization of communication without degrading performance?What abstractions are needed to allow hardware to adapt to the programmer,rather than the other way around? Can communication efficiency be improvedwhen running on an application-specific communication platform? The projectanswers these questions by exploring abstractions and algorithms to profile aparallel program's communication, synthesize a custom network design, andimplement it in a configurable network architecture substrate. The researchmethods center around the X10 language, and include compiler instrumentationpasses, offline communication profile analyses, development of a portablenetwork intermediate representation, and network place and route softwarealgorithms.

并行软件开发的每一个方面都比串行程序更复杂。这项研究的重点是复杂性的主要来源之一：应用程序内通信。目前，程序员的责任是找到他们的应用程序的通信模式到目标系统的通信基础设施的有效映射。这项研究通过开发灵活的通信架构以及相关的工具和算法来扭转这一责任，这些工具和算法允许目标平台专门用于特定的应用程序，而不是相反。除了减轻程序员的负担外，专门化还有可能提高通信效率，而自动化技术可以增加可移植性。这项研究提出了一些问题，这些问题的答案会在传统系统堆栈的几个层面上产生影响：程序员能否从特定于硬件的通信优化中解脱出来，而不是降低性能？需要哪些抽象来允许硬件适应程序员，而不是反过来？当运行在特定于应用程序的通信平台上时，是否可以提高通信效率？该项目通过探索抽象和算法来描述并行程序的通信，综合定制的网络设计，并在可配置的网络体系结构底层实现它，从而回答了这些问题。研究方法以X10语言为中心，包括编译工具遍历、离线通信轮廓分析、可移植网络中间表示的开发、网络布局和布线软件算法。