Welcome on the forums. Please post in English or French.
You are not logged in. You won't be able to post until you register and log in using the top-right bar.
I recently installed Aster 10.6 [and 11.2] on a Red Hat 6
No install problem
astout short list ok
Then i tried my usual test file and it fails like there could be numerical problem or unit problem
Same kind of files on Ubuntu LTS with aster 10.6, no issue
Could somebody check if it is running properly? [that would be a first step]
Thanks in advance
Last edited by laurent (2012-07-26 13:41:55)
I can't open your archive. Is it maybe corrupted?
A bug on the forum...
You have to rename the archive Study.tar.gz.gz
It runs successfully in 10,6, 10.7 and 11.2.
Thanks for running the case.
It means...i don't know...there is obviously a problem with the install but the 36 tests cases ran fine...
Do you know which kind of files could be useful for somebody to have a look?
I would post them on Monday and in the meantime, thanks for the help / have a nice weekend
Why not run the approx 3000 test cases over the weekend. That may give some hints as to what isn't working on your installation... just a wild shot. You could also try a different solver (MUMPS) to see if the error persists. Your .mess file shows that you're getting rigid body modes whereas my results show a first mode of approx 2100Hz. Perhaps your Multi Front solver is the problem?
1) This simple normal mode calculation is usually one of my "standard"
There should be no problem. Was created in Aster 9 and then successfully run on 10.0, 10.1, ...up to 11.2 with the same solver
So, i wasn't very keen on trying another solver
Multi-Front was failing [case already uploaded]
LDLT was taking forever to run. [had to stop it after 25minutes]
Mumps worked as expected
I am attaching the *.mess
3) Didn't ran the full tests during the weekend
For the short test, Aster 11.2 forma8a and forma11c failed, i will follow up on that
Have a nice day
Last edited by laurent (2012-07-23 11:24:41)
On Aster 11.2, i have the same behavior:
Multi-Front: Same error message
LDLT: takes much more time until i kill it
After, would you prefer me to continue investigations on 10.6 or 11.2 9 (if it is helpful for you)?
The case you're having trouble with is an Eigen analysis. We have some cluster configuration where Eigen analysis test-cases fail whereas they're working perfectly on the reference server.
What we have found so far is that it may be due to :
- faulty BLAS libraries (for example with the latest MKL there seems to be a problem)
Could you join your config.txt file ?
Yet your problem may not come from such problems. When looking at the .mess file that fails, we can see that the 6 frequencies found are near zero, e.g. rigid body motion ? Is this expected or completely different from the reference result ?
Here is the Code Aster 11.2 config file
In the short list of test cases, forma8a and forma11c failed. forma11c is clearly an eigen mode problem, so it seems it could be linked to the issue you mentioned above
My test file is a standard "product" for which we try to design it to have eigen frequency >650Hz and if possible >1200Hz
When there are numerical problems with this kind of standard product, it usually boils down to:
a. unit conversion between Catia and mesher
b. Young modulus in relation to a.
c. forgot to put the boundary conditions
d. unexpected issue from installation! (new to me)
The machine is Centos 6.0 with standard gcc 4.4.4 and standard blas/lapack
a. compiled version of Code Saturne runs Ok on that machine
[but i think Saturne incorporates its own blas?]
b. this is the first time i find these kinds of problems
c.usual working installation of Aster:
Debian 6 with Debian gcc and blas/lapack and/or Intel+MKL [laptop]
Ubuntu 10.04 LTS with gcc and blas/lapack and/or Intel+MKL [laptop]
Ubuntu 12.04 LTS with gcc and blas/lapack and/or Intel+MKL [laptop]
RedHat 5 with Intel+MKL [cluster]
RedHat 6 / Centos 6 with Intel+MKL [cluster]
-> that was the first time i tried Centos 6 + standard gcc+blas/lapack!
Have a nice day and thank for looking at it
Last edited by laurent (2012-07-24 03:09:28)
In the end, i choosed to change the solver to MUMPS for the time being.
It is not "nice" but i struggled too much to try to have a clean install with others compilers & blas/lapack
I will check Salome-Meca archive and maybe use that instead
Thanks for the help anyway.
Just a note to tell you that unfortunately your test-case runs ok on all of our supported platforms as well as on my local machine.
So we can't say more about this.
This was done with version 11.2.7.
Yes, i totally understand.
Maybe my message wasn't crystal-clear but when i wrote:
"choosed to change the solver to MUMPS for the time being.
It is not "nice" but i struggled too much to try to have a clean install with others compilers & blas/lapack"
"forma11c failed. forma11c is clearly an eigen mode problem, so it seems it could be linked to the issue you mentioned above"
my meaning was:
a. Something is wrong with some parts of my installation for LDLT and MULTI-FRONT
b. But i don't know how to solve it myself with the tools at hand (compiler or blas/lapack librarie)
c. Since MUMPS is working, i will use it until i found a way to have a better installation (for this, i have to wait for IT to come back from holidays!)
So, i put [STOPPED] in the title
But thanks for checking again and following-up many weeks later!
Thanks again and have a nice day