HFSS15: Troubleshooting for HFSS-IE Distributed Memory Simulations

A number of things can prevent distributed memory solutions from completing successfully. This section provides suggestions to debug problems.

It is often a good idea to set up and run a small simulation with two processes on a single machine before moving to a large cluster. Using a single machine will allow you to verify that HFSS and MPI are installed correctly while eliminating problems arising from remote installation, authentication and firewall settings.

Many problems occur because the MPI software cannot start and run due to authentication and firewall issues. Please check with your MPI vendor and their end user documentation for information about how to verify that authentication and firewall settings are correct.

If you are using MPI on Windows from Platform Computing you can test whether MPI will run by using the "mpidiag" utility:

• From a command prompt browse to the Platform Computing binaries located at

<HFSS_Installation_Directory>commonfluent_mpimultiportmpi_win64(win32)pcmpibin

• Enter "mpidiag -s <name_of_machine> -at" to run an authentication test.

• Run the authentication tests in both directions, i.e. both to and from all target machines. This will verify that MPI passwords and firewall settings are correct.

The tests need to be run in both directions because firewalls may allow communication in one direction but not the other.

Platform Computing's implementation of MPI on Windows requires that you enter a password on each machine in the cluster to run MPI solutions.

• From a command prompt browse to the Platform Computing binaries located at

<HFSS_Installation_Directory>commonfluent_mpimultiportmpi_win64(win32)pcmpibin

• To set the password run "mpidiag -s <name_of_machine> -cache -at" and enter the password at the prompt.

Recall that HFSS must be installed in the same directory on all machines in the cluster and that the cluster must be uniform (i.e. all Linux machines, all 32 bit Windows machines, or all 64 bit Windows machines)

Verify that the version of the third party MPI software is identical on all the machines and that it is listed in the table of supported versions above.

Verify that the machine names are correct and that all the machines can be reached on the network.