Traffic Congestion Backpressure or Packet Loss Retransmission on the Switch

If the profiling data shows that a Notify Wait task lasts for about 4 seconds, the network configuration is usually incorrect, causing packet loss and retransmission. You can use the hccn_tool to check the roce_new_pkt_rty_num field in the statistics to locate the fault.

If the value of this field increases during task execution, packet loss and retransmission occur on the network. Check the switch configuration.

Run the following command to check the statistics value:

hccn_tool -i {DeviceId} -stat -g