Configurable Network Protocol Accelerator (COPA)

IEEE Micro(2021)

引用 7|浏览5
暂无评分
摘要
Today's field-programmable gate arrays (FPGAs) offer not only programmable acceleration capabilities but also include advanced features that are on-par with a mainstream processor platform. However, rather than being deployed as autonomous acceleration nodes, they are generally deployed as second-class citizens under the control of a standard processor platform. The configurable network protocol accelerator (COPA) framework enables FPGAs to support custom inline/lookaside accelerators and attach directly to a 100-Gb/s Ethernet network. The hardware component provides the necessary networking/accelerator infrastructure. The software component supports an open standard API that enables the integrated accelerators to be used with network communication by applications/middleware. COPA enables a truly distributed FPGA accelerator model and has been implemented on variants of Stratix10 FPGAs. The COPA framework along with example accelerator IPs has been validated with microbenchmarks as well as proxy client/server benchmarks and achieves full network bandwidth.
更多
查看译文
关键词
field-programmable gate arrays offer not only programmable acceleration capabilities,mainstream processor platform,autonomous acceleration nodes,standard processor platform,configurable network protocol accelerator framework,Ethernet network,open standard API,integrated accelerators,network communication,truly distributed FPGA accelerator model,Stratix10 FPGAs,COPA framework,example accelerator IPs,network bandwidth
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要