Intel® Fortran Compiler
Build applications that can scale for the future with optimized code designed for Intel® Xeon® and compatible processors.
Announcements
FPGA community forums and blogs on community.intel.com are migrating to the new Altera Community and are read-only. For urgent support needs during this transition, please visit the FPGA Design Resources page or contact an Altera Authorized Distributor.
29285 Discussions

libifport.so.5: cannot open shared object file: No such file or directory

cras
Beginner
1,380 Views

 

I am currently attempting to perform quantum chemistry calculations and have encountered an issue when trying to run parallel processing using OpenMPI.

When performing calculations on a single node, everything works fine. However, I encounter errors when attempting to run the calculations across multiple nodes connected via SSH. The path appears to be set correctly, as I have confirmed that the following is in place:

LD_LIBRARY_PATH=/opt/intel/oneapi/compiler/2024.1/lib

I would appreciate any assistance you can provide in resolving this issue.

 

Labels (1)
0 Kudos
1 Solution
Ron_Green
Moderator
1,330 Views

I assume all the nodes in your cluster have the same image, so /opt/intel/oneapi/* exists identically across all the compute nodes, all paths the same, same version of oneapi etc.

 

I don't know OpenMPI well.  but most MPI implementations have a job launcher 'mpirun' or 'mpiexec' to launch the job.  Let's assume it's mpirun.  For the mpirun command for OpenMP look at it's options.  Almost all MPIs have something like an -env option or several variants.  One should tell the launcher to take your local environment var setting and use those on the remote node.  The default behaviour would be for the remote process for the other ranks to start only with what you have in your ~/.bashrc. So that LD_LIBRARY_PATH would not get copied.

You can check this - you can use mpirun to launch commands on the remote node.  The remote command can be 

mpirun -n 1 -host (or however you target the remote node )   env

 

to print out the remote environment vars.  Look for LD_LIBRARY_PATH.  Is it correct?  now play with mpirun options until that env shows that the remote process inherits your current env settings.  OR you can drop that export LD_LIBRARY_PATH setting in your ~/.bashrc although I don't recommend this as you are sure to forget you put it in there and someday it will break something.

View solution in original post

2 Replies
Ron_Green
Moderator
1,331 Views

I assume all the nodes in your cluster have the same image, so /opt/intel/oneapi/* exists identically across all the compute nodes, all paths the same, same version of oneapi etc.

 

I don't know OpenMPI well.  but most MPI implementations have a job launcher 'mpirun' or 'mpiexec' to launch the job.  Let's assume it's mpirun.  For the mpirun command for OpenMP look at it's options.  Almost all MPIs have something like an -env option or several variants.  One should tell the launcher to take your local environment var setting and use those on the remote node.  The default behaviour would be for the remote process for the other ranks to start only with what you have in your ~/.bashrc. So that LD_LIBRARY_PATH would not get copied.

You can check this - you can use mpirun to launch commands on the remote node.  The remote command can be 

mpirun -n 1 -host (or however you target the remote node )   env

 

to print out the remote environment vars.  Look for LD_LIBRARY_PATH.  Is it correct?  now play with mpirun options until that env shows that the remote process inherits your current env settings.  OR you can drop that export LD_LIBRARY_PATH setting in your ~/.bashrc although I don't recommend this as you are sure to forget you put it in there and someday it will break something.

cras
Beginner
1,252 Views

Thank you very much for your advice. Following your suggestions, I was able to resolve the issue.

By adding the necessary environment variable commands to the [runmpi] script, the problem with LD_LIBRARY_PATH was successfully addressed. As a result, the remote nodes now inherit the correct environment settings.

I truly appreciate your help and guidance on this matter. Your insights were invaluable in solving the problem.

Reply