SimAI-Simulation Bit Error Issue

In the gen_Topo_Template.py file, I additionally set -er 1e-3 to simulate link errors and evaluate performance under BER. However, the network sometimes fails to send NACK packets correctly, which causes the simulation to terminate prematurely and prevents me from obtaining results.

For topology generation,
python3 ./astra-sim-alibabacloud/inputs/topo/gen_Topo_Template.py -topo Spectrum-X -g 128 -gt A100 -bw 100Gbps -nvbw 2400Gbps -er 1e-3

For astra-sim-ns3,
AS_SEND_LAT=3 AS_NVLS_ENABLE=1 ./bin/SimAI_simulator -t 16 -w ./example/microAllReduce.txt -n ./Spectrum-X_128g_8gps_100Gbps_A100 -c astra-sim-alibabacloud/inputs/config/SimAI.conf

I’m not sure if it’s a problem with ReceiverCheckSeq.

Here it only checks whether the packet sequence number matches the expected sequence number, but it doesn’t check whether the packet’s tag id (which indicates which communication flow it belongs to) is consistent. This causes some cases where packet loss is not detected.

I found that in the protocol of Infiniband (https://www.afs.enea.it/asantoro/V1r1_2_1.Release_12062007.pdf), the PSN is defined by:

The PSN is defined differently from the definition of seq in the ns3 simulator. If it is defined in this way, a tag id is not required.

I hope to get your guidance and explanation on this issue！

update

I had some misunderstandings before. The problem might be that the master doesn’t have a timeout mechanism?

Reproduce the failure with gen_Topo_Template.py using the Spectrum-X command and the astra-sim-ns3 SimAI_simulator command. Inspect ReceiverCheckSeq and the master timeout behavior, then trace why NACK handling fails under -er 1e-3. Done means identifying the cause and confirming that the simulation completes and produces results under the reported configuration.