CFD Online Discussion Forums

CFD Online Discussion Forums (
-   OpenFOAM Running, Solving & CFD (
-   -   Problem with MPI? (

cwang5 May 27, 2010 07:47

Problem with MPI?
Hi guys,

I was trying to run a parallel simulation for a custom made code when I received this error message from MPI:


[1] #0  Foam::error::printStack(Foam::Ostream&)--------------------------------------------------------------------------
 An MPI process has executed an operation involving a call to the
 "fork()" system call to create a child process.  Open MPI is currently
 operating in a condition that could result in memory corruption or
 other system errors; your MPI job may hang, crash, or produce silent
 data corruption.  The use of fork() (or system() or other calls that
 create child processes) is strongly discouraged.
 The process that invoked fork was:
  Local host:          compute173 (PID 704)
  MPI_COMM_WORLD rank: 1
 If you are *absolutely sure* that your application will successfully
 and correctly survive a call to fork(), you may disable this warning
 by setting the mpi_warn_on_fork MCA parameter to 0.
  in "/home/m080031/OpenFOAM/OpenFOAM-1.6.x/lib/linux64GccDPOpt/"
 [1] #1  Foam::sigFpe::sigFpeHandler(int) in "/home/m080031/OpenFOAM/OpenFOAM-1.6.x/lib/linux64GccDPOpt/"
 [1] #2  __restore_rt at sigaction.c:0
 [1] #3  Foam::divide(Foam::Field<double>&, Foam::UList<double> const&, Foam::UList<double> const&) in "/home/m080031/OpenFOAM/OpenFOAM-1.6.x/lib/linux64GccDPOpt/"
 [1] #4  void Foam::divide<Foam::fvPatchField, Foam::volMesh>(Foam::GeometricField<double, Foam::fvPatchField, Foam::volMesh>&, Foam::GeometricField<double, Foam::fvPatchField, Foam::volMesh> const&, Foam::GeometricField<double, Foam::fvPatchField, Foam::volMesh> const&) in "/home/m080031/OpenFOAM/OpenFOAM-1.6.x/lib/linux64GccDPOpt/"
 [1] #5  Foam::tmp<Foam::GeometricField<double, Foam::fvPatchField, Foam::volMesh> > Foam::operator/<Foam::fvPatchField, Foam::volMesh>(Foam::tmp<Foam::GeometricField<double, Foam::fvPatchField, Foam::volMesh> > const&, Foam::tmp<Foam::GeometricField<double, Foam::fvPatchField, Foam::volMesh> > const&) in "/home/m080031/OpenFOAM/OpenFOAM-1.6.x/lib/linux64GccDPOpt/"
 [1] #6  Foam::incompressible::RASModels::kOmegaSST::F2() const in "/home/m080031/OpenFOAM/OpenFOAM-1.6.x/lib/linux64GccDPOpt/"
 [1] #7  Foam::incompressible::RASModels::kOmegaSST::correct() in "/home/m080031/OpenFOAM/OpenFOAM-1.6.x/lib/linux64GccDPOpt/"
 [1] #8  main in "/home/m080031/OpenFOAM/m080031-1.6.x/applications/bin/linux64GccDPOpt/flapFoam"
 [1] #9  __libc_start_main in "/lib64/"
 [1] #10  Foam::regIOobject::writeObject(Foam::IOstream::streamFormat, Foam::IOstream::versionNumber, Foam::IOstream::compressionType) const in "/home/m080031/OpenFOAM/m080031-1.6.x/applications/bin/linux64GccDPOpt/flapFoam"
 [compute173:00704] *** Process received signal ***
 [compute173:00704] Signal: Floating point exception (8)
 [compute173:00704] Signal code:  (-6)
 [compute173:00704] Failing at address: 0x2742000002c0
 [compute173:00704] [ 0] /lib64/ [0x3e1b030280]
 [compute173:00704] [ 1] /lib64/ [0x3e1b030215]
 [compute173:00704] [ 2] /lib64/ [0x3e1b030280]
 [compute173:00704] [ 3] /home/m080031/OpenFOAM/OpenFOAM-1.6.x/lib/linux64GccDPOpt/ [0x2b0d224220a1]
 [compute173:00704] [ 4] /home/m080031/OpenFOAM/OpenFOAM-1.6.x/lib/linux64GccDPOpt/ [0x2b0d208fb8e9]
 [compute173:00704] [ 5] /home/m080031/OpenFOAM/OpenFOAM-1.6.x/lib/linux64GccDPOpt/ [0x2b0d208fcc48]
 [compute173:00704] [ 6] /home/m080031/OpenFOAM/OpenFOAM-1.6.x/lib/linux64GccDPOpt/ [0x2b0d20c166a2]
 [compute173:00704] [ 7] /home/m080031/OpenFOAM/OpenFOAM-1.6.x/lib/linux64GccDPOpt/ [0x2b0d20c1b8f4]
 [compute173:00704] [ 8] /home/m080031/OpenFOAM/m080031-1.6.x/applications/bin/linux64GccDPOpt/flapFoam [0x41e8dd]
 [compute173:00704] [ 9] /lib64/ [0x3e1b01d974]
 [compute173:00704] [10] /home/m080031/OpenFOAM/m080031-1.6.x/applications/bin/linux64GccDPOpt/flapFoam(_ZNK4Foam11regIOobject11writeObjectENS_8IOstream12streamFormatENS1_13versionNumberENS1_15compressionTypeE+0xc9) [0x4199e9]
 [compute173:00704] *** End of error message ***

The custom solver ran fine for serial computation, and the current simulation ran correctly for ~1 hour before the error showed up. Anyone know what might have cause this "fork" in mpi operation? Thanks.


StSchiff July 12, 2010 09:50

same problem
Hey John,

I received the same error warning. Funny thing is, I have another case like this running and it hasn't stopped yet. The two cases differ in the choice of delta for my LES Simulation. I'm using the dynSmagorinsky model. For the case that is still running I chose delta "smooth" and for the one that stopped I chose delta "vanDriest". Other than that the cases are identical.

Did you manage to find out what caused the error?


cwang5 July 12, 2010 10:12

Hi Stefanie,

I think the cause of the error was that I was using sumMag instead of gSum for one of the calculations, resulting in a division by zero which caused the "floating point exception" error message, and crashed the solver, the crashed solver in turn caused MPI to display that error message.

Hope that helps.


StSchiff July 12, 2010 10:38

Hey John,

thanks for the fast reply. Looks like my problem is similar. In the calcDelta() function in vanDriestDelta.C there's a division that must have caused the floating point exception. For the smooth delta there is a different formular and I guess no zero division occurs here. That's why that case is still running without any trouble. I guess I'll have to rethink my bc values for nuSgs for this case.


All times are GMT -4. The time now is 19:52.