MPIEXEC and Windows

[Jack Fairchild]

Hello All and Happy New Year,

I am trying to setup a windows cluster and am having some odd anomalies. I've read many of the archives and it's clear these machines should be switched to Linux. For various reasons, that's not an option right now. I've successfully set up a windows cluster in the past (5 years ago).

All machines are dual quads.

All have the same FDS/SMV build. FDS Version 6.7.6-0

All machines have the same Windows build and updates. Windows 10 Pro for Workstations. Version 10.0.19044

I do not know the network structure.

On to the issue.

Working with just two of the machines (SMOKEY-1 and SMOKEY-2) and following section 3.2 of the user guide, Launching an FDS Job. I'm successful up to step 4. I'm using example case layer_4mesh.fds as my test case and have used both the hostfile method and just writing machine names on the command line. Same result.

At step 4 the processes are successfully assigned to the various cores on the various machines. (All core usage spikes) 4 processes, 2 machines.

"Reading FDS input file . . ." is displayed and the program stalls.

Here's the weird part. If I copy "layer_4mesh.fds" into the fds folder on the other machines the error becomes

forrtl: severe (29): file not found, unit 4, file c:\fds\layer_4mesh.smv

and these 5 files are written into the main folder on SMOKEY-1

layer_4mesh.binfo

layer_4mesh.end

layer_4mesh.sinfo

layer_4mesh.smv

layer_4mesh_git.txt

Copying layer_4mesh.smv to SMOKEY-2 allows the simulation to run, but the mesh files are split (written) between the two machines and not back to the main machine.

Suggestions, thoughts, ideas, tips?

Best,

Jack

[Glenn Forney]

Are you working with a shared directory? That is can each computer "see" the directory where you case is located?

--
You received this message because you are subscribed to the Google Groups "FDS and Smokeview Discussions" group.
To unsubscribe from this group and stop receiving emails from it, send an email to fds-smv+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/fds-smv/f01307c1-56f7-45a4-902a-3bd84dd822dan%40googlegroups.com.

[Jack Fairchild]

Hi Glenn,

Yes. C:\fds is shared with everyone on all machines and permissions are set to allow for full control, change, and read.

It is writing to both directories (machines). That is, it is writing some .sf files to one and some to the other. The files are not duplicates. They appear to correspond to the meshes being assigned to the specific machine.

Attached are snips of the two machine directories.

I tried copying all the files back to a single directory, but smokeview doesn't see them.

Best,

Jack

[Jack Fairchild]

The problem has been solved.

When entering the mpiexec command I did not include the -wdir variable. This is not stated in the user guide, it is only stated in the helpfds command.

Best,

Jack

[Kevin]

I added the -wdir to the User's Guide. If you have a chance, try the following:

mpiexec -machine hosts.txt -wdir <working directory> fds job_name.fds

where hosts.txt is a text file with

SMOKEY-1:2

SMOKEY-2:2

This means that the job requires 2+2=4 MPI processes, and SMOKEY-1 gets 2 and SMOKEY-2 gets 2.

[Jack Fairchild]

That works and assigns processes consecutively.

Process 0 and 1 go to Smokey-1. 2 and 3 go to Smokey-2.

Thank you!

Jack

[Kevin McGrattan]

Great, thanks for letting me know. I could never figure out how to do this with -hostfile.