open TELEMAC-MASCARET The mathematically superior suite of solvers

Greetings,

I am currently trying to run TELEMAC-3D (v8p4) on my university's cluster, but I am facing a major issue. The simulations in parallel do not produce the same results as in sequential mode. The same problem occurs with the provided TELEMAC-3D examples. This issue only affects 3D simulations, 2D ones work fine.

I would be very grateful for any insights on this issue.

Hi Rafia,

How different are the results? You will likely get more help if you provide specific examples and values and figures.

André Renault

First of all, thank you very much. I agree with you. Previously, I have tried to attach some figures, but each time I was blocked from publishing my question. I hope this time it worked.

The figures compare the results of a simulation I am working on: one run was done on my own computer, the second on a cluster without parallel processing, and the last one used the cluster with just two cores. I noticed that the same problem occurred even with the Telemac 3D examples. I hope the problem is now clearer. Thank you in advance.

Hi
You compare the listing information but in parallel and in sequential this is not related to the same domain so this sounds logical to see some differences.
What about the comparison of results files?

Hi,

Thank you very much for your reply. Would you mind elaborating a bit more? I didn't quite understand why it's considered a different domain, even though I'm running the exact same simulation. How does running in parallel change the domain, and how does that specifically affect the listing output? I’m asking because I didn’t encounter this issue with TELEMAC-2D — the listing results are identical in both parallel and sequential modes.

To answer your question: yes, even the results files are different when I read them — and the differences are visible even at the initial timesteps.

Hi
Parallel computation is made by domain decomposition and you will have one listing per domain. Only the one coming from the master node is printed on screen, the other are redirected to files.
If the results are different at the initial time step, this means there is another problem somewhere...

Hi,

Thank you very much for your insights. I would appreciate any suggestions you might have on how to proceed or further investigate this issue.

Hard to say how investigate precisely!
If the difference occur at the initial time step, I would try to look the splitted file and the results at T=0
Try to run your parallel simulation in dedicated directory.
If this occurs with telemac example, please indicate which one to let us check on our side but as the validation process include a comparison between results in parallel and in sequential, I'm unsure we will be able to reproduce your problem, otherwise the telemac validation will highlight a problem...

Thank you for the clarification. I’ll run some tests based on your suggestions and get back to you with more details soon.

Hello,

I have tried your suggestions, but the problem still persists. I tested several TELEMAC 3D examples, and the issue appears consistently. For instance, I ran the 'Vent' example both in sequential mode and in parallel mode using 2 cores on the cluster, and the results differ.

I’ve attached screenshots of the listing outputs, the attribute tables in BlueKenue (for an arbitrarily chosen node), and the partel_T3DGEO file from the same example.

Since these examples are part of the validated TELEMAC suite, could the issue be related to how TELEMAC-3D is installed on the cluster?