How to analyze results of SimAI-Simulation

[thunderZH963]

I have run "AS_SEND_LAT=3 AS_NVLS_ENABLE=1 ./bin/SimAI_simulator -t 16 -w ./example/microAllReduce.txt -n ./HPN_7_0_128_gpus_8_in_one_server_with_100Gbps_A100 -c astra-sim-alibabacloud/inputs/config/SimAI.conf" successfully.
But, I don't know how to analyze the results?
Can you give me some guide?

[Huoyuan100861]

I have run "AS_SEND_LAT=3 AS_NVLS_ENABLE=1 ./bin/SimAI_simulator -t 16 -w ./example/microAllReduce.txt -n ./HPN_7_0_128_gpus_8_in_one_server_with_100Gbps_A100 -c astra-sim-alibabacloud/inputs/config/SimAI.conf" successfully. But, I don't know how to analyze the results? Can you give me some guide?

In this case, two Allreduce operations within a TP were performed. You can clearly see the execution time of each communication in the "Fwd xx comm" section of the results, along with their Algobw and Busbw. These metrics are crucial indicators of collective communication performance; for more details, refer to https://github.com/NVIDIA/nccl-tests/blob/master/doc/PERFORMANCE.md.

[thunderZH963]

Where are the results from?