Performance isolation method for send queue of RDMA network interface card and RDMA network interface card

A performance isolation method for a send queue of a remote direct memory access (RDMA) network interface card, comprising:

• identifying, by the RDMA network interface card on a sending side, a tenant type based on a tenant type identifier comprised in a queue pair identifier, buffering a work queue element (WQE) of a send queue of an identified latency-sensitive tenant as a latency-sensitive WQE into a latency-sensitive group of a buffering module, and buffering a WQE of a send queue of an identified bandwidth-sensitive tenant as a bandwidth-sensitive WQE into a bandwidth-sensitive group of the buffering module, wherein at least some of to-be-processed WQEs buffered into the bandwidth-sensitive group of the buffering module are stored into a wait station, WQEs of different tenants in the latency-sensitive group and the bandwidth-sensitive group that are comprised in the buffering module are buffered based on different queues, and each node of a queue is a WQE;
• determining, by using an inter-group scheduler, whether the WQE in the latency-sensitive group or the WQE in the bandwidth-sensitive group is scheduled, sending a scheduling request to a first scheduler when it is determined that the WQE in the latency-sensitive group is scheduled, and sending the scheduling request to a second scheduler when it is determined that the WQE in the bandwidth-sensitive group is scheduled;
• scheduling, by the first scheduler, the latency-sensitive WQE in the latency-sensitive group based on the scheduling request of the inter-group scheduler by using a first scheduling algorithm, and returning a WQE scheduling result to the inter-group scheduler;
• scheduling, by the second scheduler, the bandwidth-sensitive WQE in the wait station by using a second scheduling algorithm, and returning a WQE scheduling result to the inter-group scheduler;
• transmitting, by the inter-group scheduler, the WQE scheduling results from the first scheduler and the second scheduler to a WQE processing module on the RDMA network interface card on the sending side;
• and
• processing, by the WQE processing module, the latency-sensitive WQE based on the WQE scheduling result to output all data packets corresponding to the latency-sensitive WQE, and processing the bandwidth-sensitive WQE based on the WQE scheduling result to output, based on a path maximum transmission unit, one data packet corresponding to the bandwidth-sensitive WQE, and if there is a data packet that is not output, writing a WQE corresponding to the data packet that is not output back to the wait station.