Learning Forum

[jscsgh]

Hiya, when I am running a parallel simulation in Fluent with 320,000 cell numbers, the modeling process can run well. However, when I increase the mesh number to 340,000 by just increasing the length of the domain without changing the mesh size, the following error happens: The fl process could not be started. MPI Application rank 1 exited before MPI_Finalize() with status -1073741784.

Could you please provide some solutions for this, many thanks!

[DrAmine]

340K is still small number but just check if you have proper disk space and RAM: can you handle that 340K-> do you have 1 GB ram free? If yes:
I recommend to start from the mesh, setup the case again and run!

[jscsgh]

Hiya, many thanks for your answer! Initially I thought it was the memory issue, so I instealled 96 GB of ram, the error stills happens. So I exclude the memory issue.
Another thing I noticed is that the case with 320K mesh can run with parallel+double precision, but when running the 340K case the error wil appear. However, if I changed the settings to single precision, it seems to run OK (but I am missing some key results for post processing with high accuracy). Do you have any idea what would cause this. Many thanks again.

[Rob]

Can you run a mesh check, and also post the results of Domain-Info-Size here for both cases? A screen shot of both meshes may also help.

[jscsgh]

Dear Rob, many thanks for your help. Basically the meshes are created by slicing a block with uniform grid size of 4 um. This is the mesh information for the 326,250 case (0.2 mm x 0.58 mm x 0.18 mm), which can run with parallel+double precision.
This is the mesh information for the 348,710 case (0.2 mm x 0.62 mm x 0.18 mm), only the Y dimension was increased by 0.4 mm. But it can only run with parallel+single precision. If I attempted to use double precision, error ÔÇ£PI Application rank 1 exited before MPI_Finalize() with status -1073741784.ÔÇØ would occur.
Many thanks for your time to have a look at it for me. Cheers!

[Rob]

Nothing obvious there. Read in the smaller case onto 4 cores and then read the larger mesh into that Fluent session. That (mostly) removes set up errors, or rather differences between the two models.

[jscsgh]

Dear Rob, that's exactly what I did, just replacing the smaller mesh with bigger mesh without changing anything else. But error ÔÇ£The fl process could not be started. MPI Application rank 1 exited before MPI_Finalize() with status -1073741784ÔÇØ would still happen.
What's more interesting is that if I further increase the mesh number to around 50K, even the parallel+single wouldn't run, and it shows the same error. Do you think this would be related to the UDF settings I applied?
Many thanks again for your help.

[DrAmine]

Which models are involved? Can you add some screenshots? Mesh Quality is fine. I assume mesh check won't detect anything weird.

[Rob]

What does the UDF do? You'd missed that in the OP.

[DrAmine]

Okay UDF: which UDF's are you using? Probably you are hardcoding Cell Zone ID or anything else?

[jscsgh]

Dear Dr Amine and Rob, many thanks for your help again. The UDF is about a ray-tracing algorithm for laser additive manufacturing process, which includes the calculation of laser beam discretization and laser bean reflection-based heat source. I think I will debug the UDF settings for higher cell numbers, since the issue may not lie in the memory and mesh qualities. What UDF settings do you think would have an impact on determining the double precision could be used or not, or causing an error ÔÇ£MPI Application rank 1 exited before MPI_Finalize() with status -1073741784ÔÇØ. would ÔÇ£Cell Zone IDÔÇØ have an impact on this? Many many thanks for your time! Cheers!

[Rob]

Would this UDF calculate view factors based on the surface facets by any chance?

[DrAmine]

Defition of float variables might be affected.