open TELEMAC-MASCARET The mathematically superior suite of solvers

Good morning,

I have a mesh of 242383 nodes and when I launch the calculus on TELEMAC-2D using 8 processors, I have no errors but nothing happens, the calculus is blocked on 'partitionning T2DGEO' since I launched it on friday afternoon.

I already launched a calcul with 120000 nodes on 4 processors and it works very well.

I don't think I exceeded the maximal number of nodes.

Can you please help me find a solution?
Thank you

Hello

which version are you using?
is your case working in scalar mode or no ?
Did you run it in a debug mode (keyword DEBUGGER = 1 ) ?

My best regards

Riadh

Hello,

The version is v6p0.
I'm using parallel mode.
I'm going to try the debug mode right now.

Best regards

hey Riadh,

The keyword DEBUGGER=1 doesn't work, I'm working on windows.

Hi,

What version of telemac are you using?, python/perl?, windows/linux?
In the temporary folder there should be a file called partel.log.
Could you put it on the forum?
If that file is not there you will have to rerun the partel command.
I will the information i asked before to tell you want command to run.

Hope it helps,
Yoann

hey yugi,

I'm using version v6p0 on windows
here is the partel file

Hello

The problem seems to be related to the number of island in your domain (356).
until version v6p0 the maximum number of island accepted by partel is 300.
To overcome the problem, you can edit the line (in partle.f)
IF(NILE.GT.300.OR.NFRSOL.GT.300.OR.NFRLIQ.GT.300) THEN
and increase 300 to 500 for instance
However, the best solution, you need to upgrade your version to v6p2 and in a couple of weeks to v6p3.

I hope that this is helpful

Riadh

Hello,

I can't find partel.f

I found partel.f, there is a section !partitionning

Could you tell me where to put the line:

! PARTITIONING
!
!

!======================================================================
! STEP 2 : PARTITIONING THE MESH
!
! OTHER PARTITIONING METHODS SHOULD BE USED (SCOTCH FOR EXAMPLE)
! ALL PROCESSORS PERFORM THIS TASK TO AVOID COMMUNICATION
! THE USE OF PARMETIS OR PTSCOTCH COULD BE USED FOR LARGER MESHES
! IF THERE WILL BE SOME MEMORY ALLOCATION PROBLEM
!======================================================================

ALLOCATE (EPART(NELEM2),STAT=ERR)
IF (ERR.NE.0) CALL ALLOER (LU, 'EPART')
ALLOCATE (NPART(NPOIN2),STAT=ERR)
IF (ERR.NE.0) CALL ALLOER (LU, 'NPART')
!
IF (NDP==3.OR.NDP==6) THEN
ETYPE = 1
ELSE
WRITE(LU,*) 'METIS: IMPLEMENTED FOR TRIANGLES OR PRISMS ONLY'
CALL PLANTE2(-1)
STOP
ENDIF

! WE ONLY USE METIS_PARTMESHDUAL AS ONLY THE FINITE ELEMENTS PARTITION
! IS RELEVANT HERE.
!
! IMPORTANT: WE USE FORTRAN-LIKE FIELD ELEMENTS NUMBERING 1...N
! IN C VERSION, 0...N-1 NUMBERING IS APPLIED!!!
!
NUMFLAG = 1
!
WRITE(LU,*) 'USING ONLY METIS_PARTMESHDUAL SUBROUTINE'

WRITE(LU,*) ' THE MESH PARTITIONING STEP METIS STARTS'
IF (TIMECOUNT) THEN
CALL SYSTEM_CLOCK (COUNT=TEMPS, COUNT_RATE=PARSEC)
TDEBP = TEMPS
ENDIF
CALL METIS_PARTMESHDUAL
& (NELEM2, NPOIN2, IKLES, ETYPE, NUMFLAG,
& NPARTS, EDGECUT, EPART, NPART)

WRITE(LU,*) ' THE MESH PARTITIONING STEP HAS FINISHED'
IF (TIMECOUNT) THEN
CALL SYSTEM_CLOCK (COUNT=TEMPS, COUNT_RATE=PARSEC)
TFINP = TEMPS
WRITE(LU,*) ' RUNTIME OF METIS ',
& (1.0*(TFINP-TDEBP))/(1.0*PARSEC),' SECONDS'
ENDIF

Hello

partel.f can be found in folder parallel/parallel_v6p0/source. Don't forget to recompile the folder after editing the routine.
Maybe it is worth to upgrade to version v6p2. This limitation is removed in the uploadable version on SVN.

With my best reagrads

Riadh