[skabilan]

Hi All,

I am running an incompressible transient case using icoFoam on 16 processors (Itanium 2, mpirun). The code stalls for a while (roughly 15 -20 mins) when it gets to solving the pressure equations.

GAMG: Solving for p, Initial residual = 0.0754381, Final residual = 0.00116701, No Iterations 2
time step continuity errors : sum local = 8.18748e-17, global = -3.0093e-18, cumulative = -1.55931e-16
GAMG: Solving for p, Initial residual = 0.0136584, Final residual = 0.00132829, No Iterations 1
time step continuity errors : sum local = 9.01088e-17, global = 2.14171e-19, cumulative = -1.55717e-16

Basically it takes less time to solve the same problem on one processor than in parallel.Can anyone help me with this?
Let me know if you need further information regarding the case.

Thanks in advance.

Warm Regards,
Senthil

[hjasak]

Choose other solver on the pressure equation - this one obviously does not work.

Hrv

[skabilan]

Hi Hrv,

I tried all the other available solvers for the pressure equation but have the same problem. The fastest is with GAMG on four processors.

Any suggestions or thoughts?

Regards,
Senthil

[hjasak]

Even better: a paper with some results:

CG-AMG paper

You can get one of the test cases

Droplet splash test case

and all the solvers are checked into the SVN version. Why don't you try to reporoduce the results from the paper and tell me what you see.

Please let me know,

Hrv

[skabilan]

Hi Hrv,

I have emailed you the results of the droplet splash test on our Altix Itanium machine for 1,2 and 4 processors. Sorry, it was an excel sheet and I did not know how to upload the file on to the forum.

Regards,
Senthil

[johndeas]

http://www.cfd-online.com/cgi-bin/Op...?pg=formatting

look for the attach{} keyword

JD

[skabilan]

Hi Hrv,

Here is the attachment. Droplet_splash.txt

Warm Regards,
Senthil

JD: Thanks for the link

[kar]

Hello,

I would like to ask a question about parallel: when I do mpirun --hostfile <machines> -np <nprocs> .., what is the correspondence of <machines> file entries and processor numbers? Thanks!

K.

[johndeas]

Running the case droppletSpalsh directly, strictly following README guidance I also get this error:

--> FOAM FATAL IO ERROR : Unknown symmetric matrix solver CG

Valid symmetric matrix solvers are :

4
(
ICCG
smoothSolver
PCG
GAMG
)


file: /home/flurec/commun/vk/08-03/27-dropletSplash/./system/fvSolution::pcorr a
t line 27.

From function lduMatrix::solver::New
in file matrices/lduMatrix/lduMatrix/lduMatrixSolver.C at line 78.

What is going on ?

[dmoroian]

--> FOAM FATAL IO ERROR : Unknown symmetric matrix solver CG

Are you using the development version 1.4.1-dev?

Dragos

[hjasak]

Try this:

- go to OpenFOAM-1.4.1-dev/src and tell me if there is a library called lduSolvers. If so, do

wmake libso lduSolvers

and make sure the library is compiled.

- go to OpenFOAM-1.4.1-dev/applications/solvers/multiphase/interFoam and look at the file Make/options. It should contain a line under EXE_LIBS saying -llduSolvers

if so, type wmake and make sure the executable is built.

- if you are not running OpenFOAM-1.4.1-dev, you will fail somewhere along the way. If you want to have a go with the legacy solvers, go to the test case and look for the file dropletSplash/fvSolution.oldSolvers and use that for system/fvSolution

Hope this helps,

Hrv

[johndeas]

I was using the released version (and currently checking out the svn...). Sorry for the noise, and thanks for your comments.

[johndeas]

Thank you Hrvoje.

After my dev version of svn worked, I had to add the -lldusolver to a customized solver. I would'nt have find hit quickly without your post.