null

Message Boards

Run problem on an HPC machine

KH
Kelin Hu, modified 4 Years ago.

Run problem on an HPC machine

Youngling Posts: 13 Join Date: 4/6/11 Recent Posts
Hello,

I compiled the 'trunk' version (4689) on an HPC machine and did a test run of '06_delwaq'. The running seemed no problem with normal screen outputs. But when looking into those output files, I found results were zero. I then tried my own case, and got same problem. The case finished 'successfully' but got 'zero' results.

Is there something wrong with my compilation? Attached please find the build log file, screen output file and the history output. I used 'run_delwaq.sh' to run the test.

Any suggestion is welcome.

Thanks,

Kelin
Qinghua Ye, modified 4 Years ago.

RE: Run problem on an HPC machine

Jedi Council Member Posts: 612 Join Date: 3/2/11 Recent Posts
Hi Kelin,

From the log file, I was a bit doubting you compiled the code using GNU compiler. I did observe differences between the results you got and mine, but yours are not zero, if I choose the DPTAVG_EColi, for example.

I guess there might be related to compiling.

Greetings,

Qinghua
Michel Jeuken, modified 4 Years ago.

RE: Run problem on an HPC machine

Jedi Knight Posts: 156 Join Date: 1/21/13 Recent Posts
Hi Kelin,

That is strange to see. I looked into your output file. It's not all zero's like Qinghua already noted, but also not looking very good. Have you inspected the monitoring file? I've not updated to the latest trunk version on my Linux machine, but it should look something like the attached file.

Michel
Qinghua Ye, modified 4 Years ago.

RE: Run problem on an HPC machine

Jedi Council Member Posts: 612 Join Date: 3/2/11 Recent Posts
Hi Kelin,

More specifically, the C++ compiler you used is GNU and Fortan is intel fortran 14.0.2. The MPI package you used is mvapich2.

Even though, I am still wondering if you can compile successfully, the results should be very different. Can you send me your executables or the trim, trih files of example 01 as well?

Thanks in advance,

Qinghua
KH
Kelin Hu, modified 4 Years ago.

RE: Run problem on an HPC machine

Youngling Posts: 13 Join Date: 4/6/11 Recent Posts
Hello Michel & Qinghua,

I rebuilt (./build.sh -intel14 -64bit) Delft3D on our machine by forcing 'CC=icc CXX=icpc' in build.sh when calling ./configure. Is it the right way?

The example was tested again. Unfortunately, the issue exists. The results for E. Coli bacteria at output locations were either zero or a linear increasing line.
For my own case which models salinity, sediment, algae and diatom. The model did take initial conditions from input, but got zero values starting from the second time step.

In the attachment, I packed my build.sh, three log files for building, the bin folder for executable files, trih & trim files for 01_standard, ada, hda & mon files for 06_ delwaq.


Thanks a lot.

Kelin

Attachments:

Qinghua Ye, modified 4 Years ago.

RE: Run problem on an HPC machine

Jedi Council Member Posts: 612 Join Date: 3/2/11 Recent Posts
Dear Kelin,

I don't know what goes wrong there. We tried to compile our code using Intel fortran 14.0.3 and it is no problem as well. As I understood the only difference then is the MPI package. But in this case, the MPI function is not used. So I don't know what happened there.

Or maybe you can try to compile it with the NOMPI option?

Or you may just use the compiled version from Deltares to continue your work.

All the best,

Qinghua