CFD Online Logo CFD Online URL
www.cfd-online.com
[Sponsors]
Home > Forums > OpenFOAM Programming & Development

Correct way to program a nonlinear cycle to run it in parallel

Register Blogs Members List Search Today's Posts Mark Forums Read

Like Tree1Likes
  • 1 Post By Chris Lucas

Reply
 
LinkBack Thread Tools Display Modes
Old   March 23, 2012, 09:47
Question Correct way to program a nonlinear cycle to run it in parallel
  #1
New Member
 
Davide Cagnoni
Join Date: Jul 2011
Posts: 14
Rep Power: 6
davide_c is on a distinguished road
Hi everybody

I developed my own solver for some weird electrostatics+cfd problem, and i'm now facing some bugs in it when i run the solver with MPI.

the errors i obtain ( with both 1KK and 10K elements meshes ) look like the following:

Code:
[moon02-05:8604] *** An error occurred in MPI_Recv
[moon02-05:8604] *** on communicator MPI_COMM_WORLD
[moon02-05:8604] *** MPI_ERR_TRUNCATE: message truncated
[moon02-05:8604] *** MPI_ERRORS_ARE_FATAL (your MPI job will now abort)
        --------------------------------------------------------------------------
mpirun has exited due to process rank 0 with PID 8604 on
node moon02-05 exiting without calling "finalize". This may
have caused other processes in the application to be
terminated by signals sent by mpirun (as reported here).
--------------------------------------------------------------------------
I also tried to check if the standard solver give the same problem, I ran InterFoam with both the standard parallel tutorial case damBreakingFine (around 8K elements) and with a finest one (800K elements) and they work perfectly fine. Given all this, i think i may be neglecting something in my code to empty the buffers.

The structure of the code contains at first a (quite long to converge, 30-40 iterations in the first steps) Newton cycle so every time the solving for the potential variation is done, some message has to be buffered between the various processes.

does anybody know of a way to empty those buffers after the solution (if this is the issue , which i am not 100% sure about) or had any similar experience and found out what was going on or how to overcome this???

thanks in advance guys

davide
davide_c is offline   Reply With Quote

Old   March 26, 2012, 03:35
Default
  #2
Senior Member
 
Christian Lucas
Join Date: Aug 2009
Location: Braunschweig, Germany
Posts: 198
Rep Power: 8
Chris Lucas is on a distinguished road
Hi,

If I remenber correctly, this error means that a function in one domain your solving need an information out of another.

E.g.: You have a BC with a swirl and this BC needs the face center. The problem is that if you devided your domain in subdomains (processors), only one subdomain has the needed face center. Therefore, when the other processors try to get the face center they can't find it and they crash.

Best Regards,
Christian
davide_c likes this.
Chris Lucas is offline   Reply With Quote

Old   March 26, 2012, 05:34
Default
  #3
New Member
 
Davide Cagnoni
Join Date: Jul 2011
Posts: 14
Rep Power: 6
davide_c is on a distinguished road
Christian, many thanks for the tip, but I am not using anything like that you suggest... all the BCs are uniform (at least until when my code crashes), and the only RHS for the equation is computed from other fields (always cellwise, so it should not matter) and afterwards corrected by the correctBoundaryConditions() method... The part of my code where it normally crashes looks like this

Code:
Exp = P * exp ((phi_0 - phi) / refPhi);
Exp.correctBoundaryConditions();
phi.storePrevIter ();

for (int nonOrth=0; nonOrth<=nNonOrthCorr; nonOrth++)
  {
    d_phi.storePrevIter ();
    Info << "\t\t.";

    if (
        solve (
               - fvm::laplacian(A,d_phi)
               + fvm::Sp(q*Exp/refPhi,d_phi)
               ==
               fvc::laplacian(A,phi)
               + q*Exp
               ).nIterations () == 0
        )
    break;

    d_phi.relax ();
  }

phi += d_phi;
phi.correctBoundaryConditions();
phi.relax ();
and the variation d_phi has only uniform zero value or zero gradient conditions in the phisical boundary...
Do you think i need to pass/correct someway not just the BCs but the volume fields too?
davide_c is offline   Reply With Quote

Old   March 26, 2012, 10:06
Default
  #4
Senior Member
 
Christian Lucas
Join Date: Aug 2009
Location: Braunschweig, Germany
Posts: 198
Rep Power: 8
Chris Lucas is on a distinguished road
Hi,

please check, but I guess the problem is in "phi.correctBoundaryConditions();". If you remove this code line, the solver should run.

Best Regards,
Christian
Chris Lucas is offline   Reply With Quote

Old   March 27, 2012, 09:24
Default
  #5
New Member
 
Davide Cagnoni
Join Date: Jul 2011
Posts: 14
Rep Power: 6
davide_c is on a distinguished road
Christian, thank you again, but awfully, the problem is not in that line: I added it - under suggestion - while trying to fix the problem.

I also checked if it was by any chance the GAMG solver to make problems, but it is not (PCG and PBiCG crash sameway after one more cycle step).

I think i tried with almost every thing i can change...
davide_c is offline   Reply With Quote

Reply

Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are On
Pingbacks are On
Refbacks are On


Similar Threads
Thread Thread Starter Forum Replies Last Post
Parallel run in fluent for multiphase flow apurv FLUENT 2 August 3, 2011 19:44
Parallel run of OpenFOAM in linux and windows side by side m2montazari OpenFOAM Running, Solving & CFD 5 June 24, 2011 03:26
First Parallel Run - need some help Gian Maria OpenFOAM 3 June 17, 2011 12:08
Unable to run OF in parallel on a multiple-node cluster quartzian OpenFOAM 3 November 24, 2009 14:37
How to run parallel in ICEM_CFD? Kiddo Main CFD Forum 2 January 24, 2005 09:53


All times are GMT -4. The time now is 05:18.