Convergence Problem to Simulate Natural Convection (Buoyancy Driven Flow) Simulation

[Mita Sarkar]

Hi,
 
I am running on Linux 2.6.31-21-generic (Ubuntu 9.10).

Processor: Intel(R) Core(TM)2 Duo CPU     E8500  @ 3.16GHz
Memory: 4057188
C Compiler: gcc

If you need other info, please let me know.  Please let me know the issue with the Code_Saturne preprocessor. 

Thanks and Regards

Mita

[David Monfort]

Hello Mita,

In addition to what Yvan asked for, could you also post the following items (in order to check your installation) ?
- go to the build folder of the preprocessor, if you haven't removed it, and attach the config.log file (the directory is something like ecs-2.0.0-rc1.build) - go to the install folder of Code_Saturne for the binaries (something like 2.0-rc1/bin), type ldd ./cs_preprocess and tell us the output
 
Thanks in advance,

David

[Mita Sarkar]

Hi David,

The config.log file is attached below. When I typed ldd ./cs_preprocess,  it tells me the following:

msarkar@falbala:~/opt/cs-2.0-rc1/bin$ ldd ./cs_preprocess
linux-vdso.so.1 =>  (0x00007fffb60d4000)
libbft.so.1 => /home2/msarkar/opt/cs-2.0-rc1/lib/libbft.so.1 (0x00007f8a49bc8000)
libhdf5-1.6.6.so.0 => /usr/lib/libhdf5-1.6.6.so.0 (0x00007f8a4989c000)
libpthread.so.0 => /lib/libpthread.so.0 (0x00007f8a49680000)
libz.so.1 => /lib/libz.so.1 (0x00007f8a49469000)
    libm.so.6 => /lib/libm.so.6 (0x00007f8a491e5000)
    libc.so.6 => /lib/libc.so.6 (0x00007f8a48e76000)
    libmpi.so.0 => /usr/lib/libmpi.so.0 (0x00007f8a48bcd000)
    libopen-rte.so.0 => /usr/lib/libopen-rte.so.0 (0x00007f8a48985000)
    libopen-pal.so.0 => /usr/lib/libopen-pal.so.0 (0x00007f8a48712000)
    libdl.so.2 => /lib/libdl.so.2 (0x00007f8a4850e000)
    libnsl.so.1 => /lib/libnsl.so.1 (0x00007f8a482f4000)
    libutil.so.1 => /lib/libutil.so.1 (0x00007f8a480f1000)
    /lib64/ld-linux-x86-64.so.2 (0x00007f8a49dd4000)

Please let me know if I need to do anything from my part. I have another queries, i want to simulate conjugate heat transfer after the test case is completed but I installed Syrthes after installing Saturne. so, there are not coupled, do I need to recompile Saturne again or there is other way to linked them? 

Thanks and Regards

Mita

[David Monfort]

I'm thinking of  two more advanced tests you could do:

Try to recompile the preprocessor without the HDF5 support (you only need it if you want to compile the code with MED support), with e.g. the following configure line:

cd /home2/msarkar/Code_Saturne-2.0-rc1/installer/ecs-2.0.0-rc1
./configure --prefix=/home2/msarkar/opt/cs-2.0-rc1 --with-bft=/home2/msarkar/opt/cs-2.0-rc1 --with-cgns=/home2/msarkar/opt/cgnslib-2.5.4 --without-hdf5 --without-med --with-metis=/home2/msarkar/opt/metis-4.0.1
make && make install

The idea is that the preprocessor is currently linked with MPI (OpenMPI to be precise) and I'd like you to test without linking to a MPI library. I assume MPI comes with HDF5 (otherwise that means that you have set CC=mpicc beforehand, so just unset it before configuring and compiling).
 
Second idea is to test the code under valgrind. To do this, install the valgrind tool and run it directly where your mesh is (provided you added the binaries directory of Code_Saturne to you PATH environment variable):

cd /path/to/the/mesh/directory
valgrind cs_preprocess -sc --mesh your_mesh

And send us the output, especially the different valgrind comments. This would also be better to have the preprocessor installed with debugging symbols (just add --enable-debug to the configure command line before make && make install).
 
David

[Mita Sarkar]

Hi David,

Thanks a lot for your continuous help. You suggested me to recompile my preprocessor. I would like to ask you one question, what is the reason of recompiling the preprocessor? Is it because it can not read the IDEAS .unv file or it can not support any tetrahedron mesh for 2D simulation?

Regards

Mita

[Yvan Fournier]

Hello,

The reason to recompile and reinstall the preprocessor is because we suspect an installation issue (we do not reproduce the crash on our machines), so removing unnecessary extra libraries may help to restrict the possible causes of this crash, and hopefully solve the issue.

Compiling it in debug mode will both:
reduce the optimization level, and thus reduce possible compiler-related issues give us more information in case of a crash or under Valgrind

Best regards,
 
  Yvan

[Mita Sarkar]

Hello,

I will recompile it as you suggested. I have another issue. Finally

I am able to run the computation for some times with a single prismmesh in thickness as suggested by
Alexandre on May 27. Now the issue is the computation crashed after
30000 time steps. I could not figure out the reason of this crash. On
the terminal it wrote
"Killed
Error running the calculation.
Check Kernel log (listing) and error* files for details

  ********************************************
         Error in calculation stage.
" 
But there is no error file and the listing file has been ended abruptly. The listing file is attached below.  

Moreover, what ever the simulation results I got up to 30000 time
steps that does not seems to be right. The velocity field and
temperature variations within the domain is attached here. I simulated
the same test case with other CFD software and got different results
which is compared with published results. I am trying to benchmark
Saturne, so we can use for our further studies.  I also plotted the
temperature variation at the mid height of the cavity along X axis that
is also does not look correct. In my simulation, I have a 2D square
cavity with two constant temperature vertical walls and two adiabatic
horizontal walls. Natural convection is happening inside the cavity
with radiation effects. It looks like somehow the domain is loosing
temperature.  The results are attached here though the simulation did
not reach steady state within this time steps. I am concern about the
velocity and temperature field predictions. There may something wrong
with the simulation. If this is the case, please let me know. The case
file and mesh file are also attached here.

Thanks and Regards

Mita

[Mita Sarkar]

In my previous post I could not attach the results plots. So, They are attached here. In my simulation, on left vertical boundary I specified constant temperature as 373 K and on right vertical side wall it is 323 K but the temperature profile at the mid height along X (horizontal) axis does not show that.

[Yvan Fournier]

Hello,

I finally managed to reproduce your crash when reading the unv file
(it was an old bug, but was silent with most compilers, including older
versions of GCC).

If you replace ecs-2.0/src/pre-post/ecs_pre_ideas.c with the
attached version (then recompile/reinstall), the preprocessor should
work better.

Otherwise, if your calculation indicates it was killed, this may be
due to some exterior cause (for example having used up all the alloted
time under a batch system/resource manager, killing the terminal from
which the calculation was launched if it was not launched using nohup,
...)

Best regards,

  Yvan

[Mita Sarkar]

Hi Yvan,

Thank you for providing the replacement file. I will replace that file and recompile the preprocessor as you suggested.

You told that the calculation was killed because of exterior cause. What might be the exact reason? Might it be killed due to memory use? Actually I again tried to run it and after a while it was killed for some reason.  Does it take lot of memories? 

I did not get back anything about the results yet. If you have any comments about the results, please let me know. Whatever the solution it provided that is not realistic. Results shows somehow it is loosing energy. It is not maintaining constant temperature on both vertical walls during the simulation though I specified constant temperature value.

Best Regards

Mita