[WaterHammer1985]

Hello,


I have 2 workstations connected via ethernet cable and a password-less ssh. The computers talk to each other, ping between each, and I can run decomposePar on the master and the new directories are made on the slave. However, when I try to run a simulation (damBreak tutorial), nothing happens and the program appears to "hang". There is no error message given, just complete radio silence.



Any help on troubleshooting (or how to force an error message so i can know where to start) would be greatly appreciated. Pertinent (maybe) information is below:


1. Both workstations have foam-extend-4.1 installed in the same directory (~/opt/foam/foam-extend-4.1) on ubuntu 18.04. I've installed ssh and nfs-kernel-server. The only directory that The ssh is connected to a working directory that is the same on both ~/home/pc/runCluster.



2. The "machine" file lists the ip address of the master (1st line) and slave (2nd line).

3. If I add a bogus character to the machine file (ie X to the IP address), an error is returned. This makes me think that the problem is happening after the machine file is read.

[WaterHammer1985]

Update:
I have not figured out the solution to this issue, yet.

Recap:
1. Two computer setup
2. SSH and nfs-kernel-server
3. decomposePar runs successfully, the directories are created on both the master and slave
4. interFoam starts running but then hangs with no output and doesn't appear to "crash". I'm unable to force an error message output.
4a. interFoam run command: mpirun --hostfile machines /opt/foam/foam-extend-4.1/bin/foamExec interFoam -parallel > log 2>&1

Questions:
1. Is there a way to force an error message when the run hangs?
2. Suggestions on how to troubleshoot the problem?

[fertinaz]

Hanging process probably indicates an MPI communication problem. You can add debugging directives in your controlDict to enable increased verbosity. I cannot remember what they're at the moment, but you can search that online.

However, I'd try the following steps first:

• Prior to running OF, try a simple mpi ping-pong test to make sure that you can run an MPI application using both nodes.
• Assuming that you're utilizing ethernet, set the following flag in your MPI command "--mca btl_tcp_if_include eth0".

Also, don't forget to source the OF environment on the compute node by default as well.

// Fatih

[AbnerGodfrey]

Select more lists, interesting categories other than radio station names, where do I search?
website internetradio a safe high quality option for users who want to listen to free radio stations.

[Davrus]

Water hammer, were you able to resolve this issue?
I appear to have the exact same scenario albeit running 2212 on Ubuntu 20.04.


Feels like an open mpi issue but I don't want to make assumptions. Same scenario as you in terms of host names.


Many thanks in advance,
David