parallel support with CGNS format not yet implemented

kirkrich · January 18, 2013, 00:23

I do not have access to Pointwise, but I do have access to VGrid and I know how to generate meshes with VGrid. I was hoping to generate larger, viscous 3D grids, like an airplane or a car, using VGrid and convert them to CGNS format for use with su^2. I downloaded CGNS 3.1.3, compiled it and recompiled su^2 with CGNS. FYI, I had initially compiled CGNS with HDF5, but I could not get su^2 to compile unless I compiled CGNS without HDF5 (using ADF I believe). I finally got su^2 compiled with CGNS, I tried to run the testcase/euler/wedge CGNS grid and I get 'parallel support with CGNS format not yet implemented!!'. I actually have only compiled su^2 with MPI, as I would like to run larger simulations. Can I use the serial version of the code to somehow convert CGNS format to su2 format? Is parallel CGNS support in the near future? Thanks...

mcolonno · January 18, 2013, 14:01

Thanks for your interest in SU^2. Let me try to clarify our near-term plans. With respect to compiling HDF5 vs. ADF shouldn't matter as far as SU^2 is concerned. Please post any specific error messages you get so we can make any changes to the build scripts that might be needed and get you a new versions ASAP. In SU^2 right now all the reading and writing of CGNS files is done in serial but this should not (I will confirm) impact running the solvers in parallel. This is due both to our implementation and some issues that exist with CGNS's parallel i/o that we wanted to steer clear of. That being said, that only impacts read / write times at the beginning and end of a solution and even extremely large CGNS files should not take very long (CGNS is very efficient). With respect to future plans yes: we expect to have full support for CGNS reading and writing shortly.

kirkrich · January 18, 2013, 15:54

Thanks for responding. As far as the error I get when I use CGNS built with HDF5 please see the attached file makeErrorHDF5.txt. Like I said, I can get su2 to compile with MPI, Metis, and CGNS (ADF) but when I try to run the euler wedge example, I get the previously listed message from the SU2_CFD executable at 'Read grid file information', see mpi_CGNS_SU2_CFD.txt. I tried to run SU2_DDC for 2 partitions and I get the same message, see mpi_CGNS_SU2_DDC.txt. To try and check if it was my CGNS build, I compiled su2 with no MPI and Metis. When I ran the euler wedge example it worked, see serialCGNS_SU2_CFD.txt

economon · January 18, 2013, 16:39

Hi guys, let me see if I can jump in and help clarify...

We are still building up our CGNS capabilities in SU2, as Mike explained. The HDF5 issue has been known for some time (http://adl.stanford.edu/docs/display...D/Known+Issues), and hopefully, this will be cleared up in the near future.

At the moment, we offer the ability to read CGNS mesh files in order to do just what you mention: provide support for an open, standard mesh format so that folks can get meshes running in SU2 without depending on a single piece of meshing software. While it is true that parallel support for CGNS is not online yet, we do offer a way to convert your CGNS meshes to SU2 native format using the following config option:

% Convert a CGNS mesh to SU2 format (YES, NO)
CGNS_TO_SU2= YES

Simply run SU2_CFD in serial with the following option, and a new SU2 mesh will be written. This is encouraged for the time being, as many of the python scripts and other modules, such as SU2_MDC, that are needed for design are currently geared toward the SU2 format only. Once the mesh is in native format, all features of the code will be available.

In short, I would recommend converting to the SU2 format for now, but expanded CGNS support is definitely on the way.

January 18, 2013, 00:23	parallel support with CGNS format not yet implemented	#1
kirkrich New Member Join Date: Jan 2013 Posts: 7 Rep Power: 13	I do not have access to Pointwise, but I do have access to VGrid and I know how to generate meshes with VGrid. I was hoping to generate larger, viscous 3D grids, like an airplane or a car, using VGrid and convert them to CGNS format for use with su^2. I downloaded CGNS 3.1.3, compiled it and recompiled su^2 with CGNS. FYI, I had initially compiled CGNS with HDF5, but I could not get su^2 to compile unless I compiled CGNS without HDF5 (using ADF I believe). I finally got su^2 compiled with CGNS, I tried to run the testcase/euler/wedge CGNS grid and I get 'parallel support with CGNS format not yet implemented!!'. I actually have only compiled su^2 with MPI, as I would like to run larger simulations. Can I use the serial version of the code to somehow convert CGNS format to su2 format? Is parallel CGNS support in the near future? Thanks...

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Superlinear speedup in OpenFOAM 13	msrinath80	OpenFOAM Running, Solving & CFD	18	March 3, 2015 06:36
Exporting Structured format of CGNS	shainath	Main CFD Forum	0	February 8, 2012 00:06
CFX11 + Fortran compiler ?	Mohan	CFX	20	March 30, 2011 19:56
Meshing support parallel multiple computers!	Kevin	Siemens	1	July 26, 2007 20:27
Fluent Parallel SUPPORT	MOHAMMED TOLBA	FLUENT	1	February 20, 2006 02:27

January 18, 2013, 14:01		#2
mcolonno New Member Michael Colonno Join Date: Jan 2013 Location: Stanford, CA Posts: 28 Rep Power: 13	Thanks for your interest in SU^2. Let me try to clarify our near-term plans. With respect to compiling HDF5 vs. ADF shouldn't matter as far as SU^2 is concerned. Please post any specific error messages you get so we can make any changes to the build scripts that might be needed and get you a new versions ASAP. In SU^2 right now all the reading and writing of CGNS files is done in serial but this should not (I will confirm) impact running the solvers in parallel. This is due both to our implementation and some issues that exist with CGNS's parallel i/o that we wanted to steer clear of. That being said, that only impacts read / write times at the beginning and end of a solution and even extremely large CGNS files should not take very long (CGNS is very efficient). With respect to future plans yes: we expect to have full support for CGNS reading and writing shortly.

January 18, 2013, 16:39		#4
economon Super Moderator Thomas D. Economon Join Date: Jan 2013 Location: Stanford, CA Posts: 271 Rep Power: 14	Hi guys, let me see if I can jump in and help clarify... We are still building up our CGNS capabilities in SU2, as Mike explained. The HDF5 issue has been known for some time (http://adl.stanford.edu/docs/display...D/Known+Issues), and hopefully, this will be cleared up in the near future. At the moment, we offer the ability to read CGNS mesh files in order to do just what you mention: provide support for an open, standard mesh format so that folks can get meshes running in SU2 without depending on a single piece of meshing software. While it is true that parallel support for CGNS is not online yet, we do offer a way to convert your CGNS meshes to SU2 native format using the following config option: % Convert a CGNS mesh to SU2 format (YES, NO) CGNS_TO_SU2= YES Simply run SU2_CFD in serial with the following option, and a new SU2 mesh will be written. This is encouraged for the time being, as many of the python scripts and other modules, such as SU2_MDC, that are needed for design are currently geared toward the SU2 format only. Once the mesh is in native format, all features of the code will be available. In short, I would recommend converting to the SU2 format for now, but expanded CGNS support is definitely on the way.