SDCI NET: Development of an Ultra-high Speed End-to-end Transport Stack based on the Packet Scale Paradigm.

SDCI NET：基于数据包规模范例的超高速端到端传输堆栈的开发。

• 批准号: 1127413
• 负责人: Jasleen Kaur Sahni
• 金额: $ 83万
• 依托单位: University of North Carolina at Chapel Hill
• 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: 2011-09-01 至 2016-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1127413&HistoricalAwards=false
• 关键词: SDCI; NET; Development; Ultra; Speed

End-to-end data transfer rate requirements in the computational science communities---including computational biology, physics, astronomy, medicine, and environmentalmodeling--- are soon to approach the terabit-per-second regime. Unfortunately, thestate-of-the-art in end-to-end transport protocol implementations scales to at most afew gigabit-per-seconds of single-stream steady-state throughput. This project willdevelop a 10Gbps+ Linux transport-layer stack of the novel paradigm of "packet-scalecongestion-control", which can potentially improve scalability by several orders ofmagnitude. For this, the project will: (i) develop a non ACK-clocked state machinethat replaces the TCP state machine, and (ii) develop both software-based as well ashardware-assisted high-speed implementations of precise inter-packet gap control andpacket timestamping. The implementation will be extensively tested on a 10Gbpsin-laboratory test-bed as well as on nationwide NLR paths. The project will thenpartner with up to three projects from the scientific domain, for deploying andevaluating the impact of the protocol stack on their infrastructure. The resultantimplementation is expected to be transformative in the design and implementation ofnon ACK-clocked transport protocols. It is expected to dramatically improve largehigh-speed bulk data transfers, without requiring expensive dedicated networks, thusrevolutionizing how the computational science community relies on the networkinginfrastructure.

计算科学界的端到端数据传输速率要求-包括计算生物学、物理学、天文学、医学和环境建模-很快就会接近每秒太比特的制度。不幸的是，最先进的端到端传输协议实现可扩展到最多每秒几吉比特的单流稳定状态吞吐量。该项目将开发一个10Gbps+Linux传输层堆栈，采用“数据包-规模-拥塞-控制”的新范例，潜在地可以将可伸缩性提高几个数量级。为此，该项目将：(I)开发一个非ACK时钟的状态机来取代TCP状态机，以及(Ii)开发基于软件和硬件辅助的精确数据包间间隙控制和数据包时间戳的高速实现。该实施将在10Gbpsin实验室试验台以及全国NLR路径上进行广泛测试。然后，该项目将与来自科学领域的最多三个项目合作，部署和评估协议栈对其基础设施的影响。实现结果有望对非ACK时钟传输协议的设计和实现产生革命性的影响。预计它将显著改善大规模高速批量数据传输，而不需要昂贵的专用网络，从而彻底改变计算科学界对网络基础设施的依赖。