open TELEMAC-MASCARET The mathematically superior suite of solvers

Hello,

I'm working on parallelism.
My mesh is about 750 000 nodes.
I have a model which is OK in scalar mode.
When I try in parallelism with "parallel processors = 1", it runs also.
But when I try with more than one processor, it crashes (I have smaller simulations that are working, that's why I'm thinking about the number of nodes).

The error is about partel.exe

Moreover I did this test : I did a simulation with a 25 000 nodes mesh, it worked. I used Fudaa to split each element in 4, which gives me a 100 000 nodes mesh (with actualisation of the boundary condition file), and then it crashed.

Do you know if there is a maximum number of nodes to run in parallel ?

Last point : the chapter 12 about parallelism in the User Manual V6p0 seems to be incomplete. Is there a way to have access to those missing pages ? In a previous manual, it was mentionned that for big domains, the parameter IMAX and IMAX2 might have to be changed... What do you think about that ?

Thank you,

Best Regards,

Benoît

Hello,

No, there is no limitation except the machine memory, and there is no more parameter to change. You could send your case that crashes so that we can test it here, but 700000 elements is common practice.

With best regards,

Jean-Michel

Thank you for your answer,

I work with BenoitD.

i attach to this message an example of a simple model which doesn't run on parallism mode (only 200 000 nodes !) and a copy of the Command Window with the error message.

This model runs in "normal" mode or on parallel mode with 1 processor.

We use a PC with this configuration :
Intel Core i7 CPU This email address is being protected from spambots. You need JavaScript enabled to view it. Ghz RAM 6Go (2.99 utilisable)
Windows Seven SP1 32 bits

Thank you for your help

regards

Hum, I'm afraid that the attachment did not work.

JMH

The files are too big for the forum.

You can download the files by copying this link on your browser :
www.yousendit.com/download/VnBwOU1kNmNtNEpFQlE9PQ

Bien reçu,

JMH

Hello,

Well, I tried on Linux with an ifort compiler, and on Unix with a HP compiler, it does work. This is all I can test here. I had to change "false" into "NON" in your parameter file. You can also remove the name of the parameter file from within the parameter file itself. It can drive you mad if you change the name...
The next problem you will see is that you impose a discharge through a boundary that has no water, the program complains because this would do infinite velocities. You can add some water depth in the initial conditions.
But this is probably not the reason of your initial crash, so we have to test on a similar machine, to see if it is a bug that appears only on some compilers.

With best regards, and happy Easter !

Jean-Michel Hervouet

We are always block to run our model in parallel mode.

We tried different ways to solve the problem but it still doesn't work.

The problem seems to come from Partel.exe, as the error message says (see in joint piece).

The partel.error.log file says :

"Partel error:
Partitioning into 8 subdomains with files:
T2DGEO
T2DCLI

returns with error code 40192
Probably corrupt input files."

Is somebody knows what means this 40192 error code ?

Our configuration is :
- Intel Core i7 CPU This email address is being protected from spambots. You need JavaScript enabled to view it. Ghz RAM 6Go (2.99 utilisable)
- Windows Seven SP1 32 bits
- Intel® Visual Fortran Compiler Professional Edition
11.0 for Windows

Regards

Hi
The link to the FTP site is too old. Could you send it one more time?

maybe you could join the partel.log file, whtich should be not too big and could help us to understand your problem.

Regards

You can download files (y.c. partel.error file) at this address:

www.yousendit.com/download/MEtSeVdsaTFtMEpFQlE9PQ

Regards