[Saideep]

Hi Guys;

I have a question regarding the domain decomposition for running a case on parallel using MPI.

I used the simple decomposition until now and ran into no problems for simple cases.

I am trying to have a domain in the shape of a "T". Please see the attached figure (the colors indicate allocation of cells to processors using another software - Gerris). Within that domain, I would like to allocate the mentioned zone (marked by red solid boundary) to a specific processor.

Reason: When I inject fluid through two inlets, the main physics occurs at the junction point and I would like to allocate that region (junction) to 1 processor.

Now, from my understanding this can't be done using the simple, hierarchical or scotch decomposition. The one left is the "manual" decomposition.

From the users manual I see that we can assign specific cells to a specific processor. (can anyone direct me here for an example).
However, I have too many cells in that region where cherry picking and allocating each individual cell within that region is rather impossible.

Can anyone advice me here how to proceed please.

Thanks;
SaiD

[Bloerb]

Manual decomposition needs a file that specifies which label is on which processor. I used setFields on a field i named "processor". This quite like setting the "alpha.water" fields in multiphase simulations. I set the field value to 0 1 2 3 4 with setFields... After some minor clean up (removing the boundary values and one line in the header) used that file as the input file for manual decomposition. Should work for the case you mentioned.

[Saideep]

Sorry for my late reply. Thanks for the idea.

According to what I understand it is as simple as setting the boundaries within the domain to allocate for a processor. I will try the way you mentioned and post back.

[melabbassi]

Originally Posted by Bloerb

Manual decomposition needs a file that specifies which label is on which processor. I used setFields on a field i named "processor". This quite like setting the "alpha.water" fields in multiphase simulations. I set the field value to 0 1 2 3 4 with setFields... After some minor clean up (removing the boundary values and one line in the header) used that file as the input file for manual decomposition. Should work for the case you mentioned.

Thanks a lot! This worked for me. I did not create an additional field, but copied the case folder and used one of the existing scalar fields (e.g. T) for setFields to modify.

[RL-S]

Any block shape decomposition works great with setFields. I described how to do that in this thread recently:
Manual decomposition using setFields