Hi,
I know this post is quite old, but I have the same problem. I'm trying to use Code-Saturne on a cluster with OpenMPI 1.3.2. Everything is working fine on 1 and 2 nodes, but for more nodes it blocks. I'm running CS with the --logp 1 option and if MPI-IO is set, all processes block when reading the preprocessor_output file. If MPI_IO is off (serial IO), all the processes block at the same place (see attachment with listing files). So maybe there are some differences between OpenMPI 1.3.1 and OpenMPI 1.3.2 regarding the MPI_IO implementation... I can understand that, but why doesn't it work without MPI-IO?
I'll also try to install OpenMPI 1.3.1 to see if it works. I'll keep you posted :).
The last version of Code-Saturne - 2.0.1 - what version of OpenMPI uses?
Thanks. Kind regards,
Marius
[ask] on parallel computing
Forum rules
Please read the forum usage recommendations before posting.
Please read the forum usage recommendations before posting.
-
Marius Poke
Re: [ask] on parallel computing
- Attachments
-
- cs_listing.zip
- (25.33 KiB) Downloaded 166 times
-
Yvan Fournier
Re: [ask] on parallel computing
Hello,
We mostly use OpenMPI 1.4.2 and 1.4.3, but any other version of Open MPI, MPICH2, or any other MPI implementation should work.
The log files do not tell us in enough detail where things break, but you could try
We usually do not have any problems, though unconfirmed issues have been reported recently on a machine using Open MPI 1.4.2 (with which we have not had any issues on many other machines)/
Best regards,
Yvan
We mostly use OpenMPI 1.4.2 and 1.4.3, but any other version of Open MPI, MPICH2, or any other MPI implementation should work.
The log files do not tell us in enough detail where things break, but you could try
We usually do not have any problems, though unconfirmed issues have been reported recently on a machine using Open MPI 1.4.2 (with which we have not had any issues on many other machines)/
Best regards,
Yvan