My experimental environment is two machines equipped with one Mellanox CX-5, directly connected.
If I want to get this result graph, how do I conduct my experiment? I have tried the ib_read/write_bw/lat of perftest. Is it necessary to write C++ code to achieve multi-threaded RDMA communication at the same time to get?

I want to get a graph of how RDMA throughput declines as the number of QPS increases