MPI Parallelization of incompressible thermal flow solver for natural convection problems*

We perform MPI parallel implementation of incompressible thermal flow solver for efficient numerical simulations of natural convection problems. The numerical scheme for solving the thermal flow is based on the monolithic projection technique (Pan et al., 2017, J. Comp. Phys.), decoupling of momentum and energy equations by approximate LU decompositions. The approximate factorization technique is also applied to segregate the equations in each direction. The Poisson equation for the pressure is solved using the Fourier diagonalization technique for fast computations.  The computation domain is decomposed with pencil-type blocks and the information of each subdomain is inter-exchanged using MPI_Alltoall communication. The data packing/unpacking in three-dimensional array data communication is employed based on the derived data type and sub-communicator for a better scalability. We investigate the parallel performance on simulating Rayleigh-Benard convection problems. Detailed results for parallel performance will be presented at the meeting.