[misch]

Hi,

I am new to OpenFOAM and I have some questions about using OpenFOAM. But I am not shure if the has been asked already before. Sorry, for that.

* Is there a possibilitiy/switch to get single cpu I/O (or single file I/O) for multi-cpu jobs?

* Is there a way to get some accumulated timing informations at the end of running an OpenFOAM application? I am interested in execution times of
different parts of the code (e.g. i/O, inter
process communication, initialisation, finalisation, solver). For the solver part time consumption is printed to screen after each time step. This is not what I want. I need more detailed informations.

* My last question is of more general nature and is about parallelisation technique used in FOAM. As far as I understand FOAM is parallelised using MPI (several MPI implementations are possible to use). But what about OpenMP or shared memory features in general? Are they planed for the future? Multi core architectures become more and more important, especially in high end HPC.

Thanks and best regards
Michael

[hjasak]

Hello Michael,

1) single CPU output: Yes, write just from processor zero. Field data will still be in pieces, but you can reconstruct this easily.

2) timing information: Yes, there is a bit called cpuTimeIncrement, but you will have to add the statements to various parts of the code to time them separately. This will allow you to time the bits you want.

3) OpenMP: Yes, but this has not been done. I believe OpenMP will sort out our multi-core CPU problems, but this requires surgery at bottom-level of containers and some careful work. Do you have experience with this?

Hrv

[misch]

Hi Hrv,

many thanks for your quick response.

1) single CPU ouput: is there a switch to add to
controlDict?

2) ok!

3) Concerning OpenMP: I have only some basic experience in openmp (fortran) programming but not in c/c++ (Sorry, innately I am a fortran programmer). But I know that shared memory programming or hybrid programming respectively is an important issue for the furture development of cdf codes in general. Particularly to reach good scalability on multi-core-multi-node machines.

Best
Michael