code_saturne User's Forum

Hello,

I'm using a cluster which PT-Scotch is not installed on it.
So, I'm going to do the partitioning on my PC using PT-Scotch and then transfer the generated partitioning and the mesh_output files to the cluster.

First of all, is it possible to transfer and use the partitioning files generated by PT-Scotch on a machine to another machine that there is no PT-Scotch installed on it?

If yes, then am I doing the correct procedures as below?

I have 8 MPI processes on the PC and I use 28 MPI processes on the cluster, but by using cs_partitioning_add_partitions I created the partition file for the 28 MPI processes. Now I have three files:
1- mesh_output
2- domain_number_28
3- domain_number_8

I did transfer the first two files to the cluster and set them as mesh_input file and partitioning file then I started the calculation. Is it right and enough?

Regards,
Mohammad

Hello,

Yes, this is the general idea.

In the new computation, you can indicate which pre-partiitioning directory (previous "partition_output") you are in the Performance Settings/Partitioning section in the GUI, or otherwise, in the "domain.partition_input" field in cs_user_scripts.py

You can check the run_solver.log to make sure the partition is read (if it is not found or not in the correct place, the code will continue with the default "Morton curve" partitioning, which is robust but usually (not always) leads to slightly lower partition quality and performance.

Best regards,

Yvan

Hello,

Thank you, Yvan.
I think it's working because I see the following lines:

Reading file: mesh_input
Finished reading: mesh_input

Reading file: partition_input/domain_number_28
Finished reading: partition_input/domain_number_28

Regards,
Mohammad