Partitioning and generating databases

[kenichi-tsuda]

Dear People

Recently, I have encountered some serious problem about partitioning and generate database issues.
If anybody experienced this kind of issue before and have any suggestions and comments, please let me know.
I'm using the code (developer version) that I have downloaded from this page on Feb 3, 2026.
I set the following options for the configuration:

---------------------------------------------------------
./configure FC=ifort 'FCFLAGS=-shared-intel -mcmodel=medium' MPIFC=mpiifort CC=icc 'CFLAGS=-shared-intel -mcmodel=medium' CXX=icpc 'CXXFLAGS=-shared-intel -mcmodel=medium' --enable-double-precision=yes
---------------------------------------------------------

I originally use the SCOTCH for partitioning as follows:

---------------------------------------------------------
Partitioning algorithm for decompose_mesh
choose partitioner: 1==SCOTCH (default), 2==METIS, 3==PATOH, 4==ROWS_PART
PARTITIONING_TYPE = 1

The process for partitioning based on SCOTCH ends properly.
However, after I start running xgenerate_database, the calculation looks being on the infinite loop.
This means the output_generatedatabase.txt stops following line for several hours:

---------------------------------------------------------
...setting up jacobian
mesh jacobian: min/max = 1.4287E+07 / 2.6250E+10

...indexing global points
creating ibool indexing : x min/max = -420000.0 / 360000.0
creating indirect addressing: nglob = 3213049
creating unique point locations

...resetting up jacobian in fault domains
----------------------------------------------------------

I asked the SE people in our company to check the memory usage for our cluster machine.
He found anything wrong and said that sufficient (several 10 Gb order) amount of memory is still available.

I'm wondering I forget to change any option or use different code for generating database.

On the other hand, if I use "4==ROWS_PART" for partitioning option, the xgenerate_database works proplerly.
However, running SPECFEM3D itself is not stable, i.e. the simulation blew up and stops suddenly after several hours running.
If I tried different number of processors (NPROC), it sometimes ends properly.
Then, I'm now looking for the proper combinations of NPROC for running SPECFEM3D.

Of course, it might be better to use SCOTCH instead of ROWS_PART.
However, I have no clue to solve this problem without setting the issues for generating database.

Sorry to bother you guys for this complex issues.
(If you need any more clarifications, please let me know)

If you have any recognition, comments and suggestions to be solved, please let me know.
Anything is deeply appreciated.

Regards

Kenichi Tsuda

Kenichi Tsuda, Ph.D.
Ground Vibration Engineering Group
Center for Safety and Reliability Engineering
Institute of Technology, Shimizu Corporation
Phone: +81-90-2551-8876
Fax: +81-3-3643-7260