Incompatibilities with Ubuntu 9.10 - was: Gmsh problems

Discussion about building and installing Elmer
petroo
Posts: 148
Joined: 13 Jan 2010, 19:07
Location: Aachen, Germany

Incompatibilities with Ubuntu 9.10 - was: Gmsh problems

Post by petroo »

Hello @all,

some weeks ago I reported about incompatibilities of Gmsh generated meshes with a recent 5.5.0 Elmer version.

After some more extensive tinkering around with CAELinux (= Ubuntu 8.04 LTS), Ubuntu 9.10, various Gmsh versions, recent Salomé-MECA direct download versions used on U 9.10, Salomé package in CAELinux, various Elmer versions (both SVN downloads and U 9.10 included) I come to the following, not really satisfying results:

# A recent Salomé download is working well in U 9.10
# Gmsh seems to work well also in U 9.10
# Elmer, as a part of CAELinux, is working as expected (even though not yet tested in practice with multicore processing)

But:

# Elmer 5.5.0 is packaged into Ubuntu 9.10 in a (at least partially) dysfunct manner: Geometries, both from Gmsh and Salomé, working in CAELinux as far as the mesh import is concerned, will fail to display properly in the U 9.10 Elmer, so an after-treatment of surfaces for boundary condition assortment or even the post-processing is dysfunct in this configuration. (Testing machine: Intel QuadCore with an ATI X1550 graphics card)

# Trying to install the SVN version of Elmer in U 9.10 fails, likely due to a changed Fortran compiler, or some recent incompatibilities in the Fortran parts treatment.

The practical result of these test results is:

# Either use CAELinux (U 8.04) with an outdated version of Salomé and - at least for my system - no satisfactory X11 driver resulting in a rather low screen resolution, or

# reboot the system each time an actual simulation is desired after some geometrical modification, or

# virtualize CAELinux, with presumably reduced processing speed for the most time-critical and long-lasting jobs. (not tested yet)

None of those alternatives is really convincing at the moment. :evil:

Therefore my question to the Elmer team:

# Is it likely that the Fortran originating incompatibilities may be resolved in order to re-enable an SVN compilation on a recent U 9.10 system again?

# Would it be possible to provide an updated (binary) package as an update for the U 9.10 distribution?

Kind regards,

Peter
raback
Site Admin
Posts: 4812
Joined: 22 Aug 2009, 11:57
Antispam: Yes
Location: Espoo, Finland
Contact:

Re: Incompatibilities with Ubuntu 9.10 - was: Gmsh problems

Post by raback »

Hi Peter

Perhaps it would be a quick solution if these files work with ElmerGrid to compile just that. It has fever library dependencies and should compile out of the box.

To be on the safe side, perhaps you could attach a minimal .unv and .msh (Gmsh) files and I can test that they really do work.

One problem may be that the user defines physical entities in Gmsh and does not use the '-autoclean' flag in ElmerGrid / ElmerGUI. This may create an uncontinuous numbering which will crash ElmerGUI. A good alternative is to always neglect the physical entities when saving in Gmsh. The geometric entities is good enough for Elmer and they are always defined for every entity.

-Peter
petroo
Posts: 148
Joined: 13 Jan 2010, 19:07
Location: Aachen, Germany

Re: Incompatibilities with Ubuntu 9.10 - was: Gmsh problems

Post by petroo »

Hi Peter,

thanks for your - as always - very quick response to my questions/remarks!
Perhaps it would be a quick solution if these files work with ElmerGrid to compile just that. It has fever library dependencies and should compile out of the box.
Hmm, but this would mean that I'd have to assign the properties of different BC in a rather un-visual way, right? Being not that familiar with the file contents' structures I'd very much prefer it the way it is meant to be: Opening e.g. an UNV file exported by Salomé, divide some surfaces of more complex bodies by Elmer's automatic surface disintegration and then assign BC etc. to the various boundaries.
To be on the safe side, perhaps you could attach a minimal .unv and .msh (Gmsh) files and I can test that they really do work.
I did so already in the older thread on Gmsh problems. The situation has not changed since then - I only found out a bit better where the potential errors are looming - by cross-checking with a CAELinux installation and a Salomé export.
One problem may be that the user defines physical entities in Gmsh and does not ...
After my reported tests, I don't think that the error underlying my problems is to be located in Gmsh or even in Elmer by itself, but in its coopereation with the rest of the operating system, or the included libraries therein. The geometries I tested were about the simplest possible, and the errors appeared for both Gmsh (a tetraeder) and Salomé (a brick) mesh definitions. And they did not appear if fed into the version-wise identical Elmer package (namely 5.5.0) in CAELinux.

(Btw, I tend to switching over from Gmsh to Salomé due to the latter's far superior handling and defineability of more complex geometries. Gmsh's very basic approach by the definition of individual points -> line definitions -> areas -> volumes is in practice somewhat slow in comparison to Salomé's direct definition of basic 3-d geometric objects and the possibility to even create merges, differences and such from them for object definition.)

From my perspective of a user an appropriate update of the DEB package provided for Ubuntu 9.10 would be the most favourable way to proceed, especially so for potentially lots of other users that did not make it into this forum for discussing things. Should you take this direction I'd love to support you with beta testing on my side!

Kind regards,

Peter
raback
Site Admin
Posts: 4812
Joined: 22 Aug 2009, 11:57
Antispam: Yes
Location: Espoo, Finland
Contact:

Re: Incompatibilities with Ubuntu 9.10 - was: Gmsh problems

Post by raback »

Hi Peter
petroo wrote:
Perhaps it would be a quick solution if these files work with ElmerGrid to compile just that. It has fever library dependencies and should compile out of the box.
Hmm, but this would mean that I'd have to assign the properties of different BC in a rather un-visual way, right? Being not that familiar with the file contents' structures I'd very much prefer it the way it is meant to be: Opening e.g. an UNV file exported by Salomé, divide some surfaces of more complex bodies by Elmer's automatic surface disintegration and then assign BC etc. to the various boundaries.
Well, you can always read a mesh back into ElmerGUI. In fact the ElmerGrid plugin should be (almost) the same as the stand alone ElmerGrid except that it directs the mesh into memory.

I tested further your old examples. There are following options:
  • A) Gsmh: Save a -> in Ver. 2.0 / Save all (ignore physical groups)
  • B) ElmerGUI: Mesh / Configure / elmergríd -> -autoclean
For me the current version works if either A or B is used with both examples. If neither is used there are problems. My preference is always to use A.

The .unv support of ElmerGrid was recently updated and it might perhaps be optimal to use that directly to avoid unnecessary steps. Perhaps you could also join an .unv example exported by Salome.

-Peter
petroo
Posts: 148
Joined: 13 Jan 2010, 19:07
Location: Aachen, Germany

Re: Incompatibilities with Ubuntu 9.10 - was: Gmsh problems

Post by petroo »

Hello Peter,
The .unv support of ElmerGrid was recently updated and it might perhaps be optimal to use that directly to avoid unnecessary steps. Perhaps you could also join an .unv example exported by Salome.
Of course, I am glad to do so. But again: I don't think that the problem is the import file format, but a faulty elmer package in Ubuntu 9.10 - at least with respect to the very start of a cooperation of Salomé and Elmer.

In my understanding, both CAELinux and U 9.10 take Elmer 5.5.0 as packaged distribution version. The attached screenshot shows the results of an import of the same UNV file in both U 9.10 (left side, blue window header) and a virtualized CAELinux (right side, brown window header). While the Ubuntu 8.04 based CAELinux obviously displays the geometry correctly the Elmer display running under U 9.10 is completely garbled:
unv-import.jpg
Elmer in Ubuntu 9.10 left and CAELinux right, after importing the identical UNV file, attached as well
(104.58 KiB) Downloaded 443 times
The only conclusion I am able to draw from this is a dysfunct U 9.10 package.

I'd really appreciate a resolution of this situation, since running a time-expensive computational job in a virtualized environment is somewhat sub-optimal ... :shock:

Kind regards,

Peter
Attachments
PowerStation.tar.gz
The UNV file imported in the test, as a compressed archive
(52.5 KiB) Downloaded 371 times
raback
Site Admin
Posts: 4812
Joined: 22 Aug 2009, 11:57
Antispam: Yes
Location: Espoo, Finland
Contact:

Re: Incompatibilities with Ubuntu 9.10 - was: Gmsh problems

Post by raback »

Hi Peter

The .unv file worked fine with a recent nightly build under Windows 7. Ok, there was some problem with the lower dimensional elements but still this worked:

Code: Select all

>ElmerGrid 8 2 Kraftwerk.unv -removelowdim
>ElmerGUI -> Load mesh... -> Kraftwerk
As a proof here is a solution of Laplace equation in that mesh.
Temperature distribution
Temperature distribution
kraftwerk_temp.png (29.34 KiB) Viewed 12102 times
So you could just fetch the ElmerGrid code by

Code: Select all

svn co https://elmerfem.svn.sourceforge.net/svnroot/elmerfem/trunk/elmergrid elmergrid
and compile with configure & make & make install as in
http://www.csc.fi/english/pages/elmer/sources

It should compile with any C-compiler.

-Peter
petroo
Posts: 148
Joined: 13 Jan 2010, 19:07
Location: Aachen, Germany

Re: Incompatibilities with Ubuntu 9.10 - was: Gmsh problems

Post by petroo »

Hello Peter,
... So you could just fetch the ElmerGrid code by

Code: Select all

svn co https://elmerfem.svn.sourceforge.net/svnroot/elmerfem/trunk/elmergrid elmergrid
and compile with configure & make & make install as in
http://www.csc.fi/english/pages/elmer/sources. It should compile with any C-compiler.
Did it, (after first compiling matc which was needed by elmergrid), replaced the Ubuntu 9.10 distribution binary by it and started the GUI - pitily with the same old queer results and displaying the broken mesh once again. I don't think that the fault lies in the mesh treatment itself but in the display and handling of it in ElmerGUI, but thereby effectively hindering any Elmer organized after-treatment of surfaces and boundary condition settings and such.

Did you test a fresh Ubuntu 9.10 based installation for functionality?

There is a faint chance that the effect is caused by a not very well supported graphics adapter card on my system, but I will cross-check that in the next two days, even though I do not anticipate it to be the culprit.

Kind regards,

Peter
petroo
Posts: 148
Joined: 13 Jan 2010, 19:07
Location: Aachen, Germany

Re: Incompatibilities with Ubuntu 9.10 - was: Gmsh problems

Post by petroo »

In the meantime I put a completely different graphics card into the simulation computer (changed an ATI based against an nVidia based one) and tested the ElmerGUI display of the said imported meshes: Same effect as reported before.

Accordingly there is only one explanation for me remaining:

The Elmer DEB package distributed with Ubuntu 9.10 is faulty.

It certainly would help if you installed a respective distribution on some local computer and give it a try. This is not much of an effort.


As far as my recherches have led me, there also have been changes in the handling of FORTRAN procedures that seem to have led to incompatibilities of former to recent FORTRAN routines handling. Maybe there is a connection between those two effects?

Kind regards,

Peter
hvontres
Posts: 8
Joined: 15 Feb 2010, 22:30

Re: Incompatibilities with Ubuntu 9.10 - was: Gmsh problems

Post by hvontres »

petroo wrote:In the meantime I put a completely different graphics card into the simulation computer (changed an ATI based against an nVidia based one) and tested the ElmerGUI display of the said imported meshes: Same effect as reported before.

Accordingly there is only one explanation for me remaining:

The Elmer DEB package distributed with Ubuntu 9.10 is faulty.

It certainly would help if you installed a respective distribution on some local computer and give it a try. This is not much of an effort.


As far as my recherches have led me, there also have been changes in the handling of FORTRAN procedures that seem to have led to incompatibilities of former to recent FORTRAN routines handling. Maybe there is a connection between those two effects?

Kind regards,

Peter

You might want to try using the packages from openovation instead (http://www.opennovation.org/ubuntu/) They seem to be from a later svn version than the canonical ones.
petroo
Posts: 148
Joined: 13 Jan 2010, 19:07
Location: Aachen, Germany

Re: Incompatibilities with Ubuntu 9.10 - was: Gmsh problems

Post by petroo »

Hello hvontres,
You might want to try using the packages from openovation instead (http://www.opennovation.org/ubuntu/) They seem to be from a later svn version than the canonical ones.
Tried it, but did not change a thing. ElmerGUI remains broken with respect to loading an otherwise (under CAELinux, Ubuntu 8.04) working project. I even tried to execute the solver via the GUI, but after a promising start of the mpi parallelized threads the solvers died with a segmentation fault.

I also tried the recipe given in http://tehnick-8.narod.ru/elmer/compilation.html and came somewhat closer with the compilation of a fresh SVN export under 9.10, but finally got about the same non-functioning ElmerGUI at last, showing the same symptoms as the packaged Elmer version in Ubuntu 9.10.

Thanks for your suggestions!

Peter
Post Reply