Hi [name removed],
Say you’re trying to split a relatively small mesh (30k nodes). If you split the mesh into more than 20, [name removed] will scream at you all kinds of errors.
but what if you want to have 30 vertical layers?
You end up with ‘overloaded’ cores, which each one handling more than the optimal amount of cells.
any suggestions on how to get around this issue?
[name removed]
Hi
Splitting 30k nodes mesh into more than 20 shouldn’t be a problem for telemac…
Could you detail the errors encountered in partel step?
Regards
Hi [name removed],
(I’ve compiled using Intel Fortran 2021)
In the case of n=40, I get a segmentation fault error
**forrtl: severe (174): SIGSEGV, segmentation fault occurred
**
In the case of n=25, I get the following
**
munmap_chunk(): invalid pointer
munmap_chunk(): invalid pointer
corrupted size vs. prev_size
**
In the case of n=20
no error
just for reference,
NDP NODES PER ELEMENT: 3
ELEMENT TYPE : 10
NPOIN NUMBER OF MESH NODES: 37276
NELEM NUMBER OF MESH ELEMENTS: 73344
Could it be some OS/ENV-specific problem?
Anyway, thank you!
Hello [name removed],
Would it be possible to play with your geometry file?
If you get a segmentation fault error, the classical answer is to use a build with debug options (BUILD_TYPE=Debug with the new build system), you may have more information if your computation crashes.
[name removed]
Certainly!, thank you!
calibration_30_05_22-2.zip (586 KB)
Hello [name removed],
I tried to run the following quite simple steering files for TELEMAC-2D and TELEMAC-3D for 1, 20, 25, 40, 48, 96, 192, 284, 768 and 1536 cores and they all finish correctly.
There may be an issue with the way you install openTELEMAC.
[name removed]
t2d.cas (127 Bytes)
t3d.cas (129 Bytes)
Huh, That’s weird. I took the newest installation from the git page, v9.1.
in any case, Thank you for the cross-check!
[name removed]