Meshing and solving on a supercomputer
I am in a school group trying to design ways of reducing drag on a certain object. We are trying to use the school's supercomputer to mesh and solve the models we set up.
It seems like in the help guide it explains how to set up parallel processing, but we are not able to use the GUI on the computer that we will start the job from.
Is there a way of creating a "log file" or something similar that will have the models we set up, and have this file specify the simulation to be solved?
I don't really know what I'm doing, otherwise I would try to explain this better.
I assume you don't have much experience with parallel computing with cfd. So my answer is a little bit more detailed.
A log file is just an output file.
What you're looking for is either command line options or a script.
Most "supercomputers", we call them cluster, are using a batch system to submit jobs. This makes sure, there will not be more processes running than the available number of cpus of the cluster. It also starts the next job when one is finished. You should check if your cluster uses a batch system.
Command line options for Star-ccm+ could be taken from the help file. The command line to run a parallel job is usually something like
#installationdirectoryofstarccm+/star/bin/starccm+ -batch #pathtomacro -np #numberofcpus -rsh ssh #pathtosimfile
the -rsh ssh option is only needed when there is no rsh available, depending on the cluster configuration. There might also be some MPI options be necessary, but that also depends on the cluster configuration and changes with the mpi implementation, therefore I can't write anything about this. #pathtomacro can be skipped when there is no macro to run.
Both meshing and solving on the cluster without any GUI can be made with a java macro. Especially when you want to mesh a big case on the cluster, you should also make sure it doesn't run out of memory. When using the trimmer, meshing is still a serial process (as well as the surface remesher).
But I would recommend to do the setup and meshing a PC or a usual workstation, copy the meshed sim-file to the cluster and run it there. That gives you the option to check the mesh before solving it, which is pretty much necessary to obtain good results.
And choose a number of cpus appropriate to your problem size. For exampe, it doesn't make sense to run a 500k cells case on more than 8 cpu's.
|All times are GMT -4. The time now is 16:31.|