star-cd parallel problem
I have some problems when I like to run star in the parallel mode with one node and four cpu's.
The mesh has about 2million polyhedrical cells.
I start star with the follow command:
star -dp -mpi=mpich mypc,4
star -dp -mpi=hp mypc,4
At first the calculation begins but after some time I get for both cases this message:
[ 2 ] ***Received signal 11 - EXITING
*** PROCESSOR 2 WILL STOP PROCESS NOW
MPI Application rank 1 exited before MPI_Finalize() with status 13
[ 1 ] ***Received signal 15 - EXITING
[ 4 ] ***Received signal 15 - EXITING
[ 3 ] ***Received signal 15 - EXITING
*** PROCESSOR 3 WILL STOP PROCESS NOW
*** PROCESSOR 1 WILL STOP PROCESS NOW
*** PROCESSOR 4 WILL STOP PROCESS NOW
Have you perhaps any idea which can help me?
Thank you very much for your help.
Hi,moro. I think you should not use this method for parallel. because only one PROCESSOR can get the information. you can try to use batch.
I don't see you are issuing a decomposing command.
star -dp -mpi=mpich -decompmeth=x mypc,4
-decompmeth = x : automatical decomposing
-decompmeth = s : using .set file
-decompmeth = m: manual, by user
I thought that when i write the star -command without the '-decompmeth', star will used the standard decomposition method: metis
But I don't know, is it really necessary to write?
Sorry, but I am a beginner in CFD and have not really experience with star.
Thank you for your help.
-batch Generate script for submission to batch queueing systems.
Thank you very much.
The command I posted before was for parallel computing NOT for Batch mode which requires queuing. You have to specify the command decompmeth for star to do the decomposition.
All these commands are in the Appendices section of the user guide.
All the best
When I submit a job on multiple processors I use the command:
star -dp computername,numberofprocessors
so, as example:
star -dp star01,4
This has the mpi set to default and the decomp set to default. I find that the default settings work on versions 3.26 and higher. Sometimes if you are trying to decompose a file that is too big (for example you are on a 32 bit system), you may have to decompose in a seperate step;
star -decomp,4 (I forget whether the ',' is necessary)
Then, to execute do
star -dp -nodecomp star01,4
If you are using the same geometry file for another run and only changing the problem file, then you can save a little time using the nodecomp option.
Hope this helps,
At first thank you for your answer.
Get it right, that the reason for my mpi problem is that my case is to big?
Have I the opportunity to change some parameter in the case.param file during the run of my case in star? Apply star this change directly?
Thank you very much
|All times are GMT -4. The time now is 12:40.|