CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   SU2 Installation (http://www.cfd-online.com/Forums/su2-installation/)
-   -   Windows 7 parallel calculations (http://www.cfd-online.com/Forums/su2-installation/111912-windows-7-parallel-calculations.html)

curky January 17, 2013 17:33

Windows 7 parallel calculations
 
Hi.
I'm trying to run tutorial turbulent case over the wing using SU2 on Windows 7 64bit. I installed 32bit version (Setup-Windows-x86.exe) due to 32 bit MIPCH2 instalation (ver.1.4.1.p1). For serial calcs it works fine but when I trying to run parallel it crashes. MIPCH is tested and works fine. I found out that I need to create three system variables to run script parallel_computation.py.

SU2_RUN = C:\Stanford ADL\SU2\SU2Py
SU2_HOME = C:\Stanford ADL\SU2
PATH = C:\Stanford ADL\SU2\SU2Py

All scripts, .su2 and .cfg are placed in \SU2Py folder. When I copy parallel version of executables (SU2_CFD.exe , SU2_DDC.exe) it crashes immediately:

C:\Stanford ADL\SU2\SU2Py>parallel_computation.py -f lam_NACA0012.cfg -p 2
Traceback (most recent call last):
File "C:\Stanford ADL\SU2\SU2Py\merge_solution.py", line 346, in <module>
input_file = open("%s_%s.vtk" % (volume_flow_filename, domain+1))
IOError: [Errno 2] No such file or directory: 'flow_1.vtk'

Note that I use windows command line to run this.
If I use Serial versions of SU2_CFD.exe , SU2_DDC.exe, then it run successfully but after I get covergence, program gives the same error:

IOError: [Errno 2] No such file or directory: 'flow_1.vtk'

In this case it gives 'flow.vtk'. Task manager shows that yhe calculation both cores are used in this calcs.
I belive that there may be an issues with parallel version of executables.
Can somebody help me with that?

amarkkassery January 18, 2013 11:52

Hi curky,

I have gotten the same error after playing a little bit. The issue here is that the MPICH2 library has not been installed correctly.When you try to run MPICH with a program on the command line, it just completely skips it. So that's the source of error. The merge solution is the process of merging the solution after the file has been created, however it throws that error because it has no file to merge.

I haven't gotten the MPICH2 library installed correctly as yet. Will let you know, as soon as possible.

Sorry about the last post!

mcolonno January 27, 2013 16:40

MPICH2 for Windows is not always easy to make work properly; the installation guide for Windows must be followed exactly. I would take two steps to debug this: 1) Please run one of the little test programs included with MPICH2 to confirm MPICH2 is installed and working properly. 2) Check your system for any other applications that use MPICH2. Unlike SU^2 these applications frequently include their own build of MPICH2 (or another MPI package) and the executables always have the same name. Hence, depending on the order of things in your path, you may be inadvertently trying to run SU^2 with another application's mpirun (for example). My solution to this was to rename my MPICH2 executables to something unique and add the appropriate dir to my system path. Once I did this everything worked for me in Windows with MPI. Good luck.

curky January 28, 2013 15:17

That problem was solved by installing new package of SU^2. Now I have an issue with SU2_DDC.exe. When I try to run parallel job it crashes during domain decomposition.

C:\Stanford ADL\SU2\bin>parallel_computation.py -f turb_ONERAM6.cfg -p 2
-------------------------------------------------------------------------
| SU2 Suite (Domain Decomposition Code) |
-------------------------------------------------------------------------

------------------------ Physical case definition -----------------------
Input mesh file name: mesh_ONERAM6_turb_hexa.su2

-------------------------- Output information ---------------------------
Don't visualize the partitions.

------------------- Config file boundary information --------------------
Navier-Stokes wall boundary marker(s): WING.
Far-field boundary marker(s): FARFIELD.
Symmetry plane boundary marker(s): SYMMETRY.

---------------------- Read grid file information -----------------------
Three dimensional problem.
43008 interior elements. 46417 points, and 0 ghost points.
3 surface markers.
2560 boundary elements in index 0 (Marker = FARFIELD).
1408 boundary elements in index 1 (Marker = WING).
2688 boundary elements in index 2 (Marker = SYMMETRY).

----------------------- Preprocessing computations ----------------------
Identifying vertices.

--------------------------- Read FFD information ------------------------
There is no FFD box definition. Just in case, review the .su2 file

---------------------- Performing mesh partitioning ---------------------
Finished partitioning using METIS 5.0.2. (4365 edge cuts).

job aborted:
rank: node: exit code[: error message]
0: Ziom-Komputer: -1073741571: process 0 exited without calling finalize
There is no geometry file (GetnZone))!
Press any key to exit...
There is no geometry file (GetnZone))!
Press any key to exit...


Running SU2_DDC.exe finishes without creating no file.

C:\Stanford ADL\SU2\bin>SU2_DDC.exe turb_ONERAM6.cfg
-------------------------------------------------------------------------
| SU2 Suite (Domain Decomposition Code) |
-------------------------------------------------------------------------

------------------------ Physical case definition -----------------------
Input mesh file name: mesh_ONERAM6_turb_hexa.su2

-------------------------- Output information ---------------------------
Don't visualize the partitions.

------------------- Config file boundary information --------------------
Navier-Stokes wall boundary marker(s): WING.
Far-field boundary marker(s): FARFIELD.
Symmetry plane boundary marker(s): SYMMETRY.

---------------------- Read grid file information -----------------------
Three dimensional problem.
43008 interior elements. 46417 points, and 0 ghost points.
3 surface markers.
2560 boundary elements in index 0 (Marker = FARFIELD).
1408 boundary elements in index 1 (Marker = WING).
2688 boundary elements in index 2 (Marker = SYMMETRY).

----------------------- Preprocessing computations ----------------------
Identifying vertices.

--------------------------- Read FFD information ------------------------
There is no FFD box definition. Just in case, review the .su2 file

------------------------- Exit Success (SU2_DDC) ------------------------

C:\Stanford ADL\SU2\bin>

I've tryied to change names of MPICH2 executables and run it but it gives the same results. Note that MPICH2 is properly installed and tested on provided examples. I belive that this must be SU2_DDC issue.

curky February 5, 2013 15:10

Is anybody who can help with this problem?
Is anybody who succesfully run parallel job on Win7??

mcolonno February 5, 2013 15:40

Please try running the serial version of DDC to partition your mesh (outside of the Python script) then run parallel CFD again. From the above the parallel executable is launching but the grid partitioning portion (METIS) is unhappy in parallel.
Please note that the Python scripts are experimental in Windows at this time. First confirm the basic executables are running (use the .exe files directly) and then we'll debug the Python driver script.

amarkkassery February 5, 2013 16:49

Quote:

Originally Posted by mcolonno (Post 406179)
Please try running the serial version of DDC to partition your mesh (outside of the Python script) then run parallel CFD again. From the above the parallel executable is launching but the grid partitioning portion (METIS) is unhappy in parallel.
Please note that the Python scripts are experimental in Windows at this time. First confirm the basic executables are running (use the .exe files directly) and then we'll debug the Python driver script.

Hi Mike,

I can confirm that the SU2_DDC package are not working when I tried to do what you've asked to do. The problem is that when I change my NUMBER_PART to anything but, 0 or 1, SU2_DDC just quits, while I am unable to define the NUMBER_PART on the command line. And, for NUMBER_PART = 0 and 1, SU2_DDC doesn't need to do anything.

I have tried to run the SU2_DDC included in windows x64 in serial and MPI packages.

Hope you can help.

-Abhilash

mcolonno February 5, 2013 16:55

I'll have to try and test your case side by side with mine. I can run DDC on my Windows 7 x64 system without an issue it has to be something specific to the case or Windows config itself. I will do some testing and get back to you as soon as I can.

curky February 5, 2013 17:10

I run it before... It doesn't work for parallel nor serial .exe file.
I done quick check of the surce code and it looks that the problem is in file SU2_DDC.cpp line:

if (geometry->GetnDim() == 3)

It seems that program ignores this if because I didn't get info:

--------------------------- Read FFD information ------------------------

in command line. I've look through the code and can't find where SetnDim() is called, where number of dimensions is read from file. Could you tell me if I am wrong and show place in the code?
Propably next if:

if (config->GetnDomain() > 1)

is next problem...
Unfortunately I can't debug this on my own because I can't compile it:

Ziom@Ziom-Komputer
/cygdrive/c/Stanford_ADL/trunk/SU2_DDC/config
$ make all
Compiling "/cygdrive/c/Stanford_ADL/trunk/SU2_DDC/obj/geometry_structure.o"
In file included from /cygdrive/c/Stanford_ADL/trunk/Common/src/geometry_structure.cpp:24:
/cygdrive/c/Stanford_ADL/trunk/Common/src/../include/geometry_structure.hpp:63: internal compiler error: Segmentation fault
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://cygwin.com/problems.html> for instructions.
makefile:47: recipe for target `/cygdrive/c/Stanford_ADL/trunk/SU2_DDC/obj/geometry_structure.o' failed
make: *** [/cygdrive/c/Stanford_ADL/trunk/SU2_DDC/obj/geometry_structure.o] Error 1

Much apprieciate for help.

amarkkassery February 5, 2013 17:30

Quote:

Originally Posted by curky (Post 406194)
I run it before... It doesn't work for parallel nor serial .exe file.
I done quick check of the surce code and it looks that the problem is in file SU2_DDC.cpp line:

if (geometry->GetnDim() == 3)

It seems that program ignores this if because I didn't get info:

--------------------------- Read FFD information ------------------------

in command line. I've look through the code and can't find where SetnDim() is called, where number of dimensions is read from file. Could you tell me if I am wrong and show place in the code?
Propably next if:

if (config->GetnDomain() > 1)

is next problem...
Unfortunately I can't debug this on my own because I can't compile it:

Ziom@Ziom-Komputer
/cygdrive/c/Stanford_ADL/trunk/SU2_DDC/config
$ make all
Compiling "/cygdrive/c/Stanford_ADL/trunk/SU2_DDC/obj/geometry_structure.o"
In file included from /cygdrive/c/Stanford_ADL/trunk/Common/src/geometry_structure.cpp:24:
/cygdrive/c/Stanford_ADL/trunk/Common/src/../include/geometry_structure.hpp:63: internal compiler error: Segmentation fault
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://cygwin.com/problems.html> for instructions.
makefile:47: recipe for target `/cygdrive/c/Stanford_ADL/trunk/SU2_DDC/obj/geometry_structure.o' failed
make: *** [/cygdrive/c/Stanford_ADL/trunk/SU2_DDC/obj/geometry_structure.o] Error 1

Much apprieciate for help.

I am guessing that the su2 file (mesh file) includes information whether it is 2 dimensional or 3 dimensional.

mcolonno February 5, 2013 17:36

What compiler are you trying use above?

curky February 5, 2013 17:38

I use gcc thru cygwin.

mcolonno February 5, 2013 17:40

I see. There's no specific reason why that won't work but this is not tested or support so we can't be of much help there. Our Windows builds are native, built with Visual Studio 2010 + the Intel compiler suite. I will try to do some more DDC testing to see if I can reproduce this.

curky February 5, 2013 17:45

Do you have maybe project files (Visual Studio)? I've got MSDN version of this compiler and if you could share me these files I could test it a little on my own.

economon February 20, 2013 03:35

Hi,

Is this issue still causing problems (cygwin is less tested than other platforms)? Can you please try to use Metis version 4.0.3 and compile/rerun SU2_DDC?

T

shirazbj February 20, 2013 07:29

I have the same DDC problem, see this thread:

http://www.cfd-online.com/Forums/su2...indows-xp.html

shirazbj February 23, 2013 05:02

Quote:

Originally Posted by curky (Post 406211)
Do you have maybe project files (Visual Studio)? I've got MSDN version of this compiler and if you could share me these files I could test it a little on my own.

you can try my su2_ddc.exe, here:

https://sites.google.com/site/gmsh2su2/download

curky February 26, 2013 14:34

shirazbj, I have checked your SU2_DDC.exe but haven't noticed any difference with the orginal one. Does this file works fine on your computer?

shirazbj February 26, 2013 18:51

Quote:

Originally Posted by curky (Post 410261)
shirazbj, I have checked your SU2_DDC.exe but haven't noticed any difference with the orginal one. Does this file works fine on your computer?

yes, it works on my computer. It outputs grid files for each patition. If you are not uisng the py script, you need to set NUMBER_PART= 8 in the cfg file. Change '8' to what number of threads you want depends on your cpu.

it is a serial version. The original one doesn't working because it declares a big array and that stoped the program running.

curky March 4, 2013 14:57

Yes, finally works!!! Great job. It will be good to include this solution in the next release of SU^2.


All times are GMT -4. The time now is 11:42.