CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   SU2 (http://www.cfd-online.com/Forums/su2/)
-   -   Problem with inv_ONERAM6 parallel calculation (http://www.cfd-online.com/Forums/su2/125811-problem-inv_oneram6-parallel-calculation.html)

burchio_cfd October 31, 2013 13:10

Problem with inv_ONERAM6 parallel calculation
 
1 Attachment(s)
Hi to everyone,
I have a strange problem with the oneram6 parallel calculation:
when te python script is started (with 6 cpu) all seems to work fine,
i've got no errors and the calculation arrives at convergence but looking at the system monitor during the calculation i see that only 1 core of the 6 is working. Can anyone explain this?:confused:
My system is a linux machine running Ubuntu 11.10 64 bit and the cpu is an
Intel CORE i7 3930k.
For more details there is a log of the terminal.

Thanks in advance.

fpalacios November 3, 2013 20:37

Quote:

Originally Posted by burchio_cfd (Post 460053)
Hi to everyone,
I have a strange problem with the oneram6 parallel calculation:
when te python script is started (with 6 cpu) all seems to work fine,
i've got no errors and the calculation arrives at convergence but looking at the system monitor during the calculation i see that only 1 core of the 6 is working. Can anyone explain this?:confused:
My system is a linux machine running Ubuntu 11.10 64 bit and the cpu is an
Intel CORE i7 3930k.
For more details there is a log of the terminal.

Thanks in advance.

Thanks,

It seems to me that SU2 is working properly in your computer, could you please just check that you speed up the code when increasing the number of cores? e.g. 2 cores is twice faster than 1.

Anyway, my recommendation is to use the latest version of the code that you can find in https://github.com/su2code . We have improved a lot the parallelization in the current SU2 version.

Best,
Francisco

burchio_cfd November 4, 2013 06:02

Thank you for your reply.
As your suggestion i have tried to run calculation with 1 core and 2 core, with python script and launching manually mpirun processes:
- with python script with 1 core the simulation reaches 195 iter in 10 minutes and with 2 cores in 12 minutes!!:eek:
- launching mpirun manually the time is 5 minutes.
In all cases the postprocessing indicates the correctness of the simulations.
I use python 2.7 and mpich2 1.5. However i'll try install the latest version of the code.

Thanks again!

burchio_cfd November 15, 2013 11:41

Problem solved!
I've made a fresh install of Ubuntu 13.10, installing mpich2 via apt get.
Now all work fine!:)

Combas December 2, 2013 11:27

Hello,

I am having trouble with the parallel compilation. I tried with 2 versions of openmpi but it does not work.
Could you please tell me how you did to compile with mpich2 since in the configure options (see the example below), you need to get the path to "mpicxx" and I have not found this file for mpich2.

./configure --prefix=/path/to/install/SU2 --with-CGNS-lib=/path/to/CGNS/lib --with-CGNS-include=/path/to/CGNS/headers --with-Metis-lib=/path/to/METIS/lib --with-Metis-include=/path/to/METIS/headers --with-Metis-version=5 --with-MPI=/path/to/mpicxx

Thank you in advance for your answer
Laurent

burchio_cfd December 2, 2013 12:35

Hallo Laurent,
if you are using Ubuntu, as in my case, you can install mpich2 via command
sudo apt-get install mpich2. After that you can find the mpicxx command in /usr/bin
directory. So the variable for mpi will be "--with-MPI=/usr/bin/mpicxx".

This is my command line for configure on Ubuntu 12.10:

sudo ./configure --prefix=/home/leo/sci_soft/SU2_RUN --with-CGNS-lib=/usr/local/lib --with-CGNS-include=/usr/local/include --with-Metis-lib=/home/leo/sci_soft/metis-4.0.3 --with-Metis-include=/home/leo/sci_soft/metis-4.0.3/Lib --with-Metis-version=4 --with-MPI=/usr/bin/mpicxx

I hope this will be useful for you.


NOTE: I've downgrade to 12.10 from 13.10 for Nvidia-CUDA compatibility problems.

Leonardo

Combas December 2, 2013 14:16

Thank you for your answer. The problem is that it seems to work but it does not really, since when I perform a parallel computation, it takes the same time per iteration if I do it on 2 cores or on 4 cores. In fact, when I launch a parallel computation on 4 cores, it seems to do 4 times the same thing... And I don't know if it comes from the open mpi / mpich2 or from the metis library...
I need to do some checks because I'm not sure that it really uses "mpich2" with "/usr/bin/mpicxx" or if it uses open mpi (since both are installed on my computer)

Thank you for your help
Laurent


All times are GMT -4. The time now is 02:22.