[Distributed Training on Atlas 800-9000] Ranktable file is not recognized

I am running the inception v4 training code obtained from model zoo, on Atlas 800-9000. I have some questions regarding the configuration of the distributed training (up to 8 devices on a single server). 

From the documentation, it is required to prepare a resource configuration file (the ranktable file) which should contain the NIC IP addresses of the devices available in the training server. 

However, I noticed that even if the IP addresses in the ranktable file are not valid, the training runs successfully and leverages all the 8 devices (I checked the device utilization using npu-smi and ascend-dmi tools). So my question are:

• Is the ranktable file required for distributed training on single server scenario?

• If the ranktable is not required for such a scenario, then how the training job is scheduled over the devices?