Intel MPI in Cluster: MPI_Comm_spawn hanged in case of there are two or more network interfaces

[SarahVN]

Hi Everyone,

I'm developing app which used to run in Cluster: Controller and Worker machines.
The instruction to launch App instance in Controller machine as below: 
      mpiexec -host ControllerName -n 1 -wdir C:/Agonda/WorkDir Agonda.exe

The Agonda.exe on Controller machine will pass the instruction as below to MPI_Comm_spawn method to launch 2 instances in Worker machine:

      -host WorkerName -wdir C:/Agonda/WorkDir Agonda.exe
      -host WorkerName -wdir C:/Agonda/WorkDir Agonda.exe
 
Result:

1. In case of there is ONLY one network interface in Controller and Worker machine enabled, the app is started successfully. There are two instance of Agonda.exe in Worker machine.

2. In case of there are two or more network interfaces in Controller or Worker machine enabled, the app is hanged at MPI_Comm_spawn  method. There is ONLY one instance of Agonda.exe in Worker machine.

Question:

1. Why is Intel-MPI app failed in case of there are two or more network interfaces enabled?
2. How can I resolve this issue?

Thanks
Nobita

[ShivaniK_Intel]

Hi,

Thanks for reaching out to us.

Could you please let us know the MPI version, OS details, cluster details, and interconnect you have been using?

Could you please set I_MPI_DEBUG=20 and share the debug information?

Below are the sample steps to set I_MPI_DEBUG

In Linux:

I_MPI_DEBUG=20 mpirun -n <number-of-processes> -ppn <processes-per-node> -f <hostfile> ./myprog

In Windows:

set I_MPI_DEBUG=20

mpiexec -n <# of processes> test.exe 

Thanks & Regards

Shivani

[ShivaniK_Intel]

Hi,

As we didn't hear back from you, Could you please provide the details that have been asked in my previous post so that we can investigate more on your issue?

Thanks & Regards

Shivani

[ShivaniK_Intel]

Hi,

I have not heard back from you. This thread will no longer be monitored by Intel. If you need further assistance please raise a new question.

Thanks & Regards

Shivani