On Apr 15, 2014, at 12:09 PM, Xiao, Jianjun (IKET) <[email protected]> wrote:
> Dear Barry, > > Yes, I have structured grid. > > Say, we totally have one million cells (100*100*100). 100,000 cells are solid > cells, and the shape of the solid is irregular. The other cells are fluid > cells. > > In principle, we could use one DMDA. And give the property to each cell. It > means the code knows which cell is solid and which is fluid. Then the problem > of load balancing occurs: if all the solid cells accumulates, say at a > corner, the work load will not be balanced. > > Maybe you have some other good ideas. Thank you My recommendation in this case is to use a regular grid (since you want to) but use an unstructured representation using DMPlex. This way the cells can be partitioned out among processes for load balancing anyway you want. DMDA (and most “structured grid” representation codes) don’t have support for dynamic “odd” partitioning for load balance since they wish to partition the regular grid into nice smaller regular grids. Barry > > JJ > ________________________________________ > From: Barry Smith [[email protected]] > Sent: Tuesday, April 15, 2014 6:30 PM > To: Xiao, Jianjun (IKET) > Cc: [email protected] > Subject: Re: [petsc-users] Two DMDAs for conjugate heat transfer > > DMDA are for structured grids. That is each DMDA represents a structured > grid. Do you have fluid everywhere and solid everywhere or are some cells > fluid and some cells solid? > > Barry > > > > On Apr 15, 2014, at 11:12 AM, Xiao, Jianjun (IKET) <[email protected]> > wrote: > >> Dear developers, >> >> I am writing a CFD code to simulate the conjugate heat transfer. I would >> like to use two DMDAs: one is for the fluid cells, and the other one is for >> the solid cells. >> >> Here are the questions: >> >> 1. Is it possible to have two different DMDAs for such a purpose? How the >> data in these two DMDAs communicate with each other? Are there any similar >> examples? >> >> 2. How to deal with the load balancing if DMDA is used? Or it is simply >> impossible? >> >> Thank you. >> >> Best regards >> JJ >
