Mitigating the Impact of Packet Corruption Loss in RDMA Networks with Switch Assistance

As cloud computing increases its demand for high-performance networks, RDMA networks have become a mainstream technology to support high-speed data transmission. Traditionally, RDMA networks are assumed lossless by enabling Priority Flow Control (PFC). However, packet loss due to packet corruption is still inevitable and common, which has a serious impact on the performance of RDMA networks. Therefore, this paper proposes Protego, a solution to mitigate the impact of packet corruption loss in RDMA networks. Protego leverages the switch's in-network vision and packet processing capabilities to detect packet corruption loss, and helps recovery to alleviate performance degradation. Protego considers different timeout situations caused by packet loss and provides fine-grained processing, and intercepts redundant packets in advance to save bandwidth. Protego is well compatible with the RDMA protocol, and our experiments show the effectiveness of Protego.