|
[Sponsors] |
Parallel AMR Redistribution - agglomeration maps |
|
LinkBack | Thread Tools | Search this Thread | Display Modes |
October 12, 2012, 14:40 |
Parallel AMR Redistribution - agglomeration maps
|
#1 |
Senior Member
Kyle Mooney
Join Date: Jul 2009
Location: San Francisco, CA USA
Posts: 323
Rep Power: 17 |
Howdy Foamers,
I'm currently working on adding a load balancer to run with dynamicRefineFvMesh. I feel that adding such a capability will vastly improve the capabilities of openFoam in terms of scaling and efficiency. I'm actually rather surprised that is has yet to be implemented considering dynamic adaptive mesh refinement is such a useful tool. For anyone not familiar with what I'm talking about check out the interDyMFoam damBreak tutorial in 2.1.x (I think the 1.6-ext has an issue). Currently, if you perform a re-balance mid-run, the refinementHistory data is lost and all cells are considered un-refined. By the next time step the refiner starts refining cells it shouldn't be and everything goes to heck. My question that I'm putting out there is: 1: Is anyone out there familiar with the oct-tree storage style of the 'refinementHistory' structures? At first I thought I understood it but after performing some simple 2 cell refinement studies I realize I'm missing something. 2: Has anyone worked with the agglomeration maps that are built into the 'decomposer' functions? From what I can tell, if we move a parent cell and all of its oct-tree refined children together during a balance, we can maintain the validity of the refinementHistory data. see: Foam::refinementHistory::distribute(mapDistributeP olyMesh) in OpenFOAM/src/dynamicMesh/lnInclude/refinementHistory.H in OF2.1.x. If we can work with the refinementHistory correctly, we can build the agglomeration map, and force Scotch to keep the cells together. I know there are a lot of people out there that need to be able to run large scale AMR simulations. I feel like if just a few of us stick our heads together for a few days we'll be able to set this up and really expand and accelerate our work. Cheers!! Kyle P.S. Sorry I spelled agglomeration wrong in the title of the thread... |
|
|
|
Similar Threads | ||||
Thread | Thread Starter | Forum | Replies | Last Post |
Script to Run Parallel Jobs in Rocks Cluster | asaha | OpenFOAM Running, Solving & CFD | 12 | July 4, 2012 22:51 |
AMR and parallel run | lichmaster | OpenFOAM | 6 | May 3, 2012 07:23 |
CFX parallel multi-node jobs fail w/ SLURM on Ubuntu 10.04 | danieru | CFX | 0 | February 17, 2012 06:20 |
HP MPI warning...Distributed parallel processing | Peter | CFX | 10 | May 14, 2011 06:17 |
Parallel Computing Classes at San Diego Supercomputer Center Jan. 20-22 | Amitava Majumdar | Main CFD Forum | 0 | January 5, 1999 12:00 |