A Survey of RoCEv2 Congestion Control

The traditional TCP/IP protocol stack cannot meet the high-bandwidth and low-latency requirements of data center applications for the network. Due to the kernel bypass and zero-copy features of RoCEv2 (RDMA on Converged Ethernet v2), the RoCEv2 protocol stack has been widely deployed in high-speed networks of data centers. The large-scale deployments of RoCEv2 network depends on the lossless network based on PFC (Priority Flow Control) protocol. However, PFC will lead to congestion diffusion, deadlock and other performance problems. Therefore, RoCEv2 network needs an effective congestion control mechanism to avoid network congestion to ensure high-performance transmission. This paper briefly introduces the principle of RoCEv2, the target of RoCEv2 congestion control and the problems and challenges it faces. It also classifies the congestion control protocols, analyzes the main research in the field of RoCEv2 congestion control in recent years, and finally looks forward to the development of RoCEv2 congestion control in the future, pointing out the direction for future research work.