Design of Converged Network Coding Layer for the Ethernet and HPC High-Speed Network

摘要

With the rapid development of the data center, high-performance computing (HPC), and multi-server system, the relevant applications present different requirements and characteristics. The data transmission is difficult between their interconnection networks, with the urgent demand for the highly converged network adapting to different interconnection networks. Hence, the multi-network convergence becomes an essential trend in the development of the interconnection network. The work proposed a converged network coding sublayer for the Ethernet network and HPC high-speed networks. Firstly, based on the low latency and high-reliability requirements of the HPC high-speed network, we analyzed the relationship between the parameters and the latency of RS code and proposed a new RS(272,257) code for the HPC high-speed network with low latency and high reliability. Then, based on the RS(528,514) code used in 100 Gbps Ethernet and the new RS(272,257) code, the converged interconnection network coding layer was designed. Finally, the work simulated the latency performance of the RS(528,514) code and the new RS(272,257) code and assessed the resource consumption of the converged network coding sublayer, the RS(528,514) coding sublayer, and the RS(272,257) coding sublayer. By the comprehensive assessment of error correction capability, latency performance, and resource consumption, the new RS(272,257) code was an ideal forward errorcorrection code to meet the requirements of the HPC highspeed interconnection network. The converged coding sublayer, configured to support the Ethernet or the HPC high-speed network, was implemented at the cost of finite resources reusing common logic of RS(272,257) and RS(528,514) as much as possible.