Domain Decomposition

ertan · August 30, 2009, 03:18

Hi all,
I've been working on parallelization of a 3D FVM based CFD code running on a backward staggered grid with two ghost cells at the end of each direction. I use MPI for massage passing.

I wrote the boundary exchange subroutines, and I believe they are working fine. However, I have doubts about the necessity of solver based modifications. To be more clear, do I need to make changes to the solver, should I leave it as it is?

I have some experience with PVM and parallelization of Finite Difference based CFD code. I remember, we made no midifications to the solver, and it was working just fine. Though we were using LSODES, ROWMAP, etc. I don't know if they handle those things by themselves.

I would really deeply appreciate if could find a chance of communicating with experienced fellas.

Thanks

Ertan

ignat · August 31, 2009, 05:11

Quote:

Originally Posted by ertan

Hi all,
I've been working on parallelization of a 3D FVM based CFD code running on a backward staggered grid with two ghost cells at the end of each direction. I use MPI for massage passing.

I wrote the boundary exchange subroutines, and I believe they are working fine. However, I have doubts about the necessity of solver based modifications. To be more clear, do I need to make changes to the solver, should I leave it as it is?

I have some experience with PVM and parallelization of Finite Difference based CFD code. I remember, we made no midifications to the solver, and it was working just fine. Though we were using LSODES, ROWMAP, etc. I don't know if they handle those things by themselves.

I would really deeply appreciate if could find a chance of communicating with experienced fellas.

Thanks

Ertan

If your code is fully explicit no solver modifications are needed.

ertan · September 1, 2009, 13:22

Hi Ignat,

thanks for the reply. For the time being, I run the code fully explicitly, though I will have to get it running fully implicit and Crank-Nicholson as well. If you have experience in parallelization on overlapping staggered grid, and if you don't mind, may I ask you some questions about the data exchange at the boundaries. It seems the, with the way I do, mass is not conserved.

Or if you can suggest any source worth for reading, I will appreciate that too.

Thanks

August 30, 2009, 03:18	Domain Decomposition	#1
ertan New Member Ertan Karaismail Join Date: Apr 2009 Posts: 17 Rep Power: 17	Hi all, I've been working on parallelization of a 3D FVM based CFD code running on a backward staggered grid with two ghost cells at the end of each direction. I use MPI for massage passing. I wrote the boundary exchange subroutines, and I believe they are working fine. However, I have doubts about the necessity of solver based modifications. To be more clear, do I need to make changes to the solver, should I leave it as it is? I have some experience with PVM and parallelization of Finite Difference based CFD code. I remember, we made no midifications to the solver, and it was working just fine. Though we were using LSODES, ROWMAP, etc. I don't know if they handle those things by themselves. I would really deeply appreciate if could find a chance of communicating with experienced fellas. Thanks Ertan

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Default domain	jaho	CFX	5	February 16, 2014 01:00
CFX - domain decomposition. Urgent!!!!	Elena Saldaeva	CFX	4	June 30, 2008 08:18
CFX Solver Memory Error	mike	CFX	1	March 19, 2008 08:22
rotating domain in rotating domain, different axis	Robert Stringer	CFX	3	December 4, 2006 08:04
Domain decomposition	rajesh	Main CFD Forum	2	August 31, 1999 05:22

September 1, 2009, 13:22		#3
ertan New Member Ertan Karaismail Join Date: Apr 2009 Posts: 17 Rep Power: 17	Hi Ignat, thanks for the reply. For the time being, I run the code fully explicitly, though I will have to get it running fully implicit and Crank-Nicholson as well. If you have experience in parallelization on overlapping staggered grid, and if you don't mind, may I ask you some questions about the data exchange at the boundaries. It seems the, with the way I do, mass is not conserved. Or if you can suggest any source worth for reading, I will appreciate that too. Thanks