open TELEMAC-MASCARET The mathematically superior suite of solvers

Hi,

I managed to get rid of the problem with the max. number of boundaries in parallel thanks to Clemens (post www.opentelemac.org/index.php/kunena/16-...-v7-1?start=10#20910 ).

However, I seem to have a problem with a boundary node in parallel (single core is running), but this particular node Looks pretty normal in the mesh and in the boundary file and I have no idea what is wrong….

For your information: It is a large floodplain with a lot of solid boundaries, which represent the buildings inside the area. The node is on the egde of one of these rectangular islands in the middle of the domain.



Happy for any hint,
Gabi

Hi,

I found the Problem in partel.f starting with line 964:

! Checking that there is not an error in the boundaries
IF(J.NE.NPTFR_P(I)) THEN
WRITE(LU,*) 'ERROR IN BOUNDARIES J=',J
WRITE(LU,*) 'WHILE NPTFR_P(I)=',NPTFR_P(I)
WRITE(LU,*) 'WHILE L=',L
CALL PLANTE(1)
STOP
ENDIF

Why is J not equal to NPTFR_P(I) in this case? Is NPTFR_P(I) in parallel still the number of the boundary node in the boundary file? I suppose yes.
Why is then J bigger than NPTFR_P(I) (the number of the boundary nodes correspond with the number of lines in the boundary file).

I attached the boundary file.

Best regards,
Gabi

Sorry, I didn't managed to get the file in here. Maybe it is too big?

Hi,

The test here is because in the Serafin format you have information on the boundaries both in the Serafin file (array ipobo) and in the boundary file.

The test here is that J is the number of boundary points on the subdomain I using the boundary file and NPTFR_P(I) is the number of boundary points using the array ipobo (irand in partel.F code).

So the problem is within your Serafin file that has more boundary points than the Boundary file.

Hope it helps

Hi Yugi,

thanks for the explanation, but why is the file running single core? Is this not checked there? And how to get rid of this error in a mesh with more than 2mil nodes and approx. 6000-7000 boundaries?

Many thanks,
Gabi

Hello,

I agree that this is strange, it could be a problem in the partitioner itself when there are 6000-7000 boundaries, this could exceed some maximum in the dimensions...

Regards,

Jean-Michel Hervouet

Hi Yugi and Jean-Michel,

I did some tests and found out, that the difference between J and NPTFR_P(I) is depending on the number of cores used in parallel. More nodes, greater difference...

I also checked the slf and the boundary file. The difference between J and NPTFR_P(I) are more than 700 nodes in case of 4 processors; I should thus be able to find at least one (!) of these 700 nodes in the mesh.
I attached a picture of the mesh with the bc2-file on it, but I am not able to find any mistake (the bc2 and cli file are made by BK from the slf file as usual...).




Has never anyone had a similar problem? Flood plain modelling with a lot of affected buildings?

Best regards,
Gabi

Picture was to big first..

Hi all,

to close the topic:
I solved the problem with the help of Clemens (Thanx!!).
Apparently the problem were too much overstressed elements (caused by using the mesh from another 2D software), which made problems in parallel.

Regards,
Gabi