Fatal error in MPI_Bsend when meshing
Moderator: oliveroxtoby
Forum rules
and Helpful information for the FEM forum
and Helpful information for the FEM forum
Fatal error in MPI_Bsend when meshing
Hello there,
I have this problem where I want to mesh a domain in cfMesh which has a lot of mesh refinements and I search for a good quality mesh. Thus, I want to take advantage of parallel cores so I wanted to use 10 cores to mesh my domain.
My surprise was when this Fatal error in MPI_Bsend: Invalid buffer pointer came in. I googled it (https://www.cfd-online.com/Forums/su2-i ... error.html) and found out that it has to do with the MAX_MPI_BUFFER size. It seems that, as it can be noted in the attached file, my total buffer size is only about 20MB.
How do I increase MAX_MPI_BUFFER size? I run on Windows 10 Pro and have 128GB of RAM.
Thank you so much for your help.
Marc.
I have this problem where I want to mesh a domain in cfMesh which has a lot of mesh refinements and I search for a good quality mesh. Thus, I want to take advantage of parallel cores so I wanted to use 10 cores to mesh my domain.
My surprise was when this Fatal error in MPI_Bsend: Invalid buffer pointer came in. I googled it (https://www.cfd-online.com/Forums/su2-i ... error.html) and found out that it has to do with the MAX_MPI_BUFFER size. It seems that, as it can be noted in the attached file, my total buffer size is only about 20MB.
How do I increase MAX_MPI_BUFFER size? I run on Windows 10 Pro and have 128GB of RAM.
Thank you so much for your help.
Marc.
- Attachments
-
- MPI_error.txt
- (9.56 KiB) Downloaded 40 times
- oliveroxtoby
- Posts: 837
- Joined: Fri Dec 23, 2016 9:43 am
- Location: South Africa
Re: Fatal error in MPI_Bsend when meshing
Try this: After writing the mesh case, click 'Edit' and modify the Allmesh file. Add the line
Code: Select all
export MPI_BUFFER_SIZE=200000000
Please report whether this works, and I'll make it the default.
Re: Fatal error in MPI_Bsend when meshing
Hi Oliver,
Thanks for your reply.
I added the line you said (attached the Allmesh file):
Altough not an error message about MPI Buffer has appeared, the meshing process exited with error, I atach the log (Meshing exited with error.txt).
Marc.
Thanks for your reply.
I added the line you said (attached the Allmesh file):
Altough not an error message about MPI Buffer has appeared, the meshing process exited with error, I atach the log (Meshing exited with error.txt).
Marc.
- Attachments
-
- Allmesh.rar
- (720 Bytes) Downloaded 29 times
-
- Meshing exited with error.txt
- (10.12 KiB) Downloaded 36 times
- oliveroxtoby
- Posts: 837
- Joined: Fri Dec 23, 2016 9:43 am
- Location: South Africa
Re: Fatal error in MPI_Bsend when meshing
Hello Oliver,
Sorry for the late reply. Here I attach the freshly written meshCase file in CfMesh.
At last I ended up meshing using gmesh tetahedral without errors and good quality mesh. If you need anything else, please just ask.
Altough, if we could fix the MPI Buffer issue would be great.
Thanks in advance.
Marc.
Sorry for the late reply. Here I attach the freshly written meshCase file in CfMesh.
At last I ended up meshing using gmesh tetahedral without errors and good quality mesh. If you need anything else, please just ask.
Altough, if we could fix the MPI Buffer issue would be great.
Thanks in advance.
Marc.
- oliveroxtoby
- Posts: 837
- Joined: Fri Dec 23, 2016 9:43 am
- Location: South Africa
Re: Fatal error in MPI_Bsend when meshing
On a machine with 16GB of RAM I get a similar error, but on a 64GB machine it goes through fine and I end up with a ~20M cell mesh. I can't explain why it fails on your 128GB machine, but it looks like something memory related.
Incidentally if you have the latest update of cfMesh, you will probably find that it runs faster with multithreading rather than with multiprocessing.
Re: Fatal error in MPI_Bsend when meshing
I will give it another try with another geometry by increasing the MPI Buffer and will reply with results.oliveroxtoby wrote: ↑Thu Jun 09, 2022 7:38 pm On a machine with 16GB of RAM I get a similar error, but on a 64GB machine it goes through fine and I end up with a ~20M cell mesh. I can't explain why it fails on your 128GB machine, but it looks like something memory related.
Please, could you explain how to change from multithreading to multiprocessing in CfdOF? Do I have to change the "number of threads" to 0 and "number of processes to 1?oliveroxtoby wrote: ↑Thu Jun 09, 2022 7:38 pm Incidentally if you have the latest update of cfMesh, you will probably find that it runs faster with multithreading rather than with multiprocessing.
Thanks,
Marc.
- oliveroxtoby
- Posts: 837
- Joined: Fri Dec 23, 2016 9:43 am
- Location: South Africa
Re: Fatal error in MPI_Bsend when meshing
Thanks for the help Oliver!