topic Hi Greg, in Intel® MPI Library

MPI error based on host order

Greg_G_ — Mon, 30 May 2016 07:29:16 GMT

I'm setting up new MPI environment with two windows 8 pc.

Running the command below in ether order works on both machines.

mpiexec -n 1 -host machine1 hostname : -n 1 -host machine2 hostname

Then running the next command below works depending on the order. test.exe is in a newly created folder on the C drive of each machine.

mpiexec -n 1 -host machine1 test.exe : -n 1 -host machine2 test.exe

When listing machine1 first and machine2 second and running the command works on either machine. Then when the hosts list is reversed the command fails on on both machines with the following error.

Thanks in advance

Hi Greg,

Artem_R_Intel1 — Mon, 30 May 2016 08:21:40 GMT

Hi Greg,

Could you please check the firewall status for both nodes? If it's enabled, try to disable it or add the appropriate permissions for MPI executables.

Hello Artem,

Greg_G_ — Mon, 30 May 2016 08:56:00 GMT

Hello Artem,

Thanks for the reply. I have added port exceptions and application exceptions to both computers firewall. But as a test I also tried disabling both computers firewalls and the behavior was not changed.

Seems strange the command can be run from either computer for a given host ordering. Then creates an error from both for the opposite order.

-Greg

Hi Greg,

Artem_R_Intel1 — Mon, 30 May 2016 09:03:18 GMT

Hi Greg,

Which version of Intel MPI Library do you use ('mpiexec -V')?

Could you please try to run your scenarios (both ones: successful and failed) with the following environment variables and provide the output:

set I_MPI_DEBUG=6

set I_MPI_HYDRA_DEBUG=on

I copied the results from the

Greg_G_ — Mon, 30 May 2016 09:43:00 GMT

I copied the results from the cmd window. Seems my window has a limit to how much is kept in the scrollable part. If there is anything missing let me know. I tried to edit the comment to include the command I used but the links started multiplying each other.

works: mpiexec -n 2 -ppn 1 -hosts machine1,machine2 test.exe
crashes: mpiexec -n 2 -ppn 1 -hosts machine2,machine1 test.exe

Hi Greg,

Artem_R_Intel1 — Mon, 30 May 2016 13:53:18 GMT

Hi Greg,

As far as I see for some reasons there're the following IP addresses detected for your nodes:

Core 192.168.56.1

Eths 192.168.200.2

Are they from the same network? If no could you please try to run your scenarios with the IP addresses specified instead of the hostnames (use the IP addresses from the same subnet).

Thanks, you found the issue.

Greg_G_ — Mon, 30 May 2016 14:04:25 GMT

Thanks, you found the issue.
They are both on the same network but one of the computers had a VirtualBox installed on them giving them two IP addresses (another user installed it).
Entering the main IP instead of the host names worked.