mpiexec is ignoring the no. of processes after each hostname in the hostfile

[YaDev]

Running MPI using a hostfile like this 

mpiexec -f hostfile executable

The hostfile has the content, for example

host1:4

host2:2

but MPI just runs the maximum number of processes on each host. It just completely ignore the process numbers in the hostfile??

[TobiasK]

@YaDev 
You are using the hostfile option while you need the machine file option. A hostfile does not contain the number of slots available, only the machinefile contains this information.

https://www.intel.com/content/www/us/en/developer/articles/technical/controlling-process-placement-with-the-intel-mpi-library.html

mpirun -machinefile hostfile executable

[YaDev]

Oh ok thank you!

[YaDev]

Hello again,

Would Intel not consider instead of having this -machinefile flag, just merge it into the hostfile under the -f flag just like what MPICH is doing? There is no reason to have this separate -machinefile flag, right?

With just the -f hostfile to list

Example 1)

host1

host2

host3

...

Example 2)

host1:3

host2

host3:5

...

The -machinefile is just quite redundant IMHO. 

[TobiasK]

Hi,
I was not aware that machinefile and hostfile are the same for MPICH, however, other implementations also have this differentiation. I will bring you feedback to the developers but I do not expect that we change the behavior due to backwards combability issues.

[YaDev · Accepted solution]

Understandable that this is low priority. It is just a suggestion but thanks!