Question about petsc4Foam performance
Dear zampini,
Hi:) I am a UCL student who is working on the performance optimization of openFoam based on petsc. When I tried to use petsc4Foam to replace the original openFoam solvers in cavity case, I found that the solving speed of petsc solver was much slower than the original one. However, in your essay: A library to plug in PETSC into the OpenFOAM framework, the petsc method should be better than the openFoam one. So I am wondering did I set any parameters wrong? And could you give me any idea about the performance of petsc4foam? I thought petsc should run faster than the original solver even in 1 node. Thanks a lot for your time!
BlockMesh: 100x100x100 ControlDict: deltaT: 0.001, endTime:0.5 Tolerance for P: 1e-04 relTol : 0 maxIter: 3000
Parameters for petsc+foam: P: bjacobi +icc+cg U: DILU+PBiCGStab
Parameters for original openFoam: P: FDIC+pcg U: symGaussSeidel+smoothSolver
Environment: 1 node: AMG 7742 with 128 cores. GCC: 9.2.0 MPI: OpenMPI 4.0.3
Thanks, Xiaofeng Fu