open TELEMAC-MASCARET The mathematically superior suite of solvers

Hello,

I am trying to run a case with more than one million nodes in parallel mode with the version V6.0

Here is the error :

****************************************
* FIN DE L'ALLOCATION DE LA MEMOIRE : *
****************************************

LECLIM : ERREUR DE LECTURE DANS LE FICHIER
DES CONDITIONS AUX LIMITES
A LA LIGNE 2787

CAUSES POSSIBLES :

1) RETOURS CHARRIOTS WINDOWS SUR UNIX ?
2) INFORMATIONS POUR LE PARALLELISME MANQUANTES ?



PLANTE : ARRET DU PROGRAMME APRES ERREUR
SORTIE DE PVM : APPEL DE P_EXIT

SORTIE DE MPI
______________________________

Indeed, when I look in the file T2DCLI00001-00000 (generated by partel in the temporary folder), I observe that nodes are written on 6 caracters :
995772 1 -1 -1 -1 -1 -1 -1 -1 -1
997631 1 -1 -1 -1 -1 -1 -1 -1 -1
999605 1 -1 -1 -1 -1 -1 -1 -1 -1
999642 1 -1 -1 -1 -1 -1 -1 -1 -1
****** 1 -1 -1 -1 -1 -1 -1 -1 -1
****** 1 -1 -1 -1 -1 -1 -1 -1 -1
****** 1 -1 -1 -1 -1 -1 -1 -1 -1
****** 1 -1 -1 -1 -1 -1 -1 -1 -1

What parameter should i increase to solve this problem ?

Thank you for any help,

Benoît D.

Hi
First of all you may try to use the version 6.1.
Maybe this problem had been removed...

All the partitionning is done in the partel program so you should probably change this program in order to solve this problem

Hope this helps

Hi,

Yes, partel.f has been corrected recently in version 6.1 and should be OK now. If not just write a new post and we'll do it.

Regards,

JMH

Hi,

I tried with version 6.1
I observe the same stars in the BC file when the node number exceeds one million :
999605 1 -1 -1 -1 -1 -1 -1 -1 -1
999642 1 -1 -1 -1 -1 -1 -1 -1 -1
****** 1 -1 -1 -1 -1 -1 -1 -1 -1
****** 1 -1 -1 -1 -1 -1 -1 -1 -1

Do you have another suggestion ?

Regards,

benoît

Hi
You should check if partel is really the lastest version.
The writing of BCfile if formatted by the variable FMT4. You could make a search for this name in partel.f
On my version, I have I7 for the first value so you could write number up to 10 milions-1.

Hope this helps

Hi,

This was the problem indeed.
I can now run meshs over 1 million nodes.

Thanks !

benoît