[Seph_01]

Hi All,

I'm having issues getting a 4 x 32 core job off the ground. Once submitted I get an MPI Error - File Name Too Long

I feel there may be a file or variable that needs tweaking somewhere. I'm using the version of MPI shipped with StarCCM 2020 and running this all on Windows Server 2019.

If I run this job at 4 x 16 then it runs fine. So MPI itself in addition to networking between the nodes is all good.

Just wondering if anyone has come across this.

Thanks

[bluebase]

There might be an issue with the plattform mpi on windows, since you are not the first one to report it:

Running on the cluster: machine file -> Filename too long???
Running on 4 node error MPI Errors[320798736]

The same is reported in this post:
Running Star on separate machines

However, this member explained he/she used intel mpi instead, which also comes with StarCCM. This avoided the issue.
Have a look into the manual, and look for the Command-Line Reference. There should be an option like "-mpi <driver>" where <driver> needs to be selected.

If this fixes your issue, please share it here.

Best,
Sebastian

[cwl]

Originally Posted by bluebase

There might be an issue with the plattform mpi on windows, since you are not the first one to report it:

Running on the cluster: machine file -> Filename too long???
Running on 4 node error MPI Errors[320798736]

The same is reported in this post:
Running Star on separate machines

However, this member explained he/she used intel mpi instead, which also comes with StarCCM. This avoided the issue.
Have a look into the manual, and look for the Command-Line Reference. There should be an option like "-mpi <driver>" where <driver> needs to be selected.

If this fixes your issue, please share it here.

Best,
Sebastian

Yes, "this member" (me) has spent lots of time monitoring commands that Star runs and trying to figure out what was wrong. And it was Platform MPI.

Installing, setting up Intel MPI:

<installdir>\mpi\intel\<installdir>\win64\bin\mpivars.bat
<installdir>\mpi\intel\<installdir>\win64\bin\hydra_service.exe -install
<installdir>\mpi\intel\<installdir>\win64\bin\hydra_service.exe -start
<installdir>\mpi\intel\<installdir>\win64\bin\mpiexec -register

Running the last command will ask you for user/password.
If user is the same as current one - just hit enter when asked for

account (domain\user) [ComputerName\UserName]:

Checking:

<installdir>\mpi\intel\<installdir>\win64\bin>mpiexec -hosts <Computer1> -validate cmd

If you still see command promt (cmd) - type whoami and hit enter, it will show which machine you're logged in now.

First, try starting cmd one the local machine with its name instead of <Computer1>,
then - the name of another machine in the network.

To use Intel MPI when starting Star-CCM+ - right in the very bottom of the New (or Load Simulation) dialog where you see something like:

starccm+ -server -on Machine1:28,Machine2:28

correct it manually to:

starccm+ -mpi intel -server -on Machine1:28,Machine2:28

PS I'm actually very confused/disappointed - as that problem still persists 4 years after my first post. It cannot be that no one has ever encountered it never reported.