STTR Phase I: Splintered Topologically Close-Packed (TCP) Offload Engine for Grid Computing and Bandwith-Delay Product (BWDP)

这个小型企业技术转让（STTR）第一阶段研究项目解决了具有极端带宽延迟产品的网络的挑战，例如美国研究机构和瑞士类似研究机构之间的40Gbps网络链路。网络中的批量数据传输需要提供所需的端点资源，以确保以经济高效的方式实现高性能。该研究项目适用于数据速率从10 Gbps扩展到40 Gbps的高性能网络，并提供与当前和未来版本的GridFTP的兼容性。此外，该研究还通过使用一种新的卸载引擎来解决这些挑战，该引擎基于最新出版物实现了TCP的分裂实现。与完全卸载方法相比，分裂TCP使用部分卸载和真正的零拷贝传输来控制NPU上所需的资源。在本项目中，卸载引擎架构将使用低成本，高性能的FPGA子系统进行研究，将派生和模拟。使用NPU开发工具包，splinter性能将被测试，以确定哪些卸载是最佳的。从这个研究项目的成功结果将显着推进卸载引擎在网格计算中使用的最新技术。直接的应用包括满足电子科学社区对网格计算的可扩展卸载引擎的需求，这需要适应具有多通道10 Gbps到40 Gbps链路的系统，同时支持现有和未来版本的GridFTP。其他应用包括使用我们的超高速Splintered TCP卸载引擎进行网格和集群计算，利用我们的开源固件。其他产品应用包括TeraGrid、网格集群WAN和刀片服务器。

This Small Business Technology Transfer (STTR) Phase I research project addresses the challenges of networks with extreme bandwidth delay products such as a 40Gbps network link between a US research facility and a similar research facility in Switzerland. Bulk data transfers in the networks need to be provided the endpoint resources required to ensure high performance in a cost effective manner. This research project accommodates such high performance networks, as data rates scale from 10 Gbps to 40 Gbps, and also provide compatibility with present and future versions of GridFTP. Additionally, the research addresses the challenges through the use of a novel offload engine with a splintered implementation of TCP based on the latest publications. In contrast to full offload approaches, splintered TCP uses partial offload and true zero-copy transfers to control the resources needed on the NPU. During this project, the offload engine architecture will be investigated using low-cost, high-performance FPGA subsystems that will be derived and simulated. Using an NPU development kit, the splinter performance will be tested to determine which offloads are optimal.Successful results from this research project will significantly advance the state of the art for offload engines used in grid computing. Immediate applications include accommodating the e-science community's need for scalable offload engines for grid computing, that is required to accommodate systems with multiple channel 10 Gbps links through 40 Gbps, while supporting present and future versions of GridFTP. Other applications include the use of our ultra high-speed Splintered TCP offload engines for Grid and cluster computing, utilizing our open source firmware. Additional product applications include the TeraGrid, Grid Cluster WANs, and blade servers.