ibv_post_send is slow in ping-pong

I tried to measure how much time it takes for each ibv_post_send (IB_WR_RDMA_WRITE) in the default rping program.

I used clock_gettime to measure and the results show that every ibv_post_send function takes around 170~180 nanoseconds. I expected it to be faster. Does any one have ideas on how to tune this? What could be the affecting factors? Many thanks in advance.