[liu]

I wrote a new solver to deal with coupled problem. There are two meshes to represent the two domains which is coupled through one common boundary patch. In one processor, it's running ok now except the speed.
I want to move to supercomputers. Can some one give me some suggestions? I am reading the code which is regarding to the parallel part in the original code. But some hint will be very helpful.

After I finish this, I am happy to release the new solver code.

[mattijs]

Your problem will be the communication across your coupled patch. The easiest solution is to make your parallel decomposition such that the whole coupled patch remains on one processor. In that way you don't have to change your code at all (probably)

Have a look a the 'manual' option to decomposePar. It reads a labelList from the constant/ (or system/ ?) directory and uses that to decompose the case. The labelList gives for every cell the processor it should go to.

You'll have to write some code to write this file so the cells around your patch are on the same processor.

2] Alternatively you can do the communication across your patch yourself. Find some examples in the utilities that use OPstream and IPstream.

3] For both method it helps if you fully work out what exactly has to be communicated between processors.

(all this is not trivial)