A Scalable Immersed Boundary Method for Multiphase Flow Simulation

We developed a highly scalable immersed boundary method (IBM) algorithm for multiphase flow simulation. In particular, a Double Bin Ghost Particle (DBGP) algorithm is designed to obtain the distributed data storage and scalable data transferring features for the particle phase. The proposed algorithm uses a new Queen/Worker data structure to indicate the particle-level and marker-level quantities and communications. In the DBGP algorithm, each particle is represented by a queen marker and surface worker markers. The queen marker determines the motion and total force of a particle. The worker marker performs the fluid-particle interaction, including velocity interpolation and force projection. A double binning system is determined through the Cartesian binning of the physical domain to relate each MPI rank with its overlapping bins. By searching the bin-to-rank map, all remote MPI ranks influenced by the local particle can be readily identified. The ghost queen marker is generated in the remote MPI rank based on the local queen marker and the queen binning structure, and same for the ghost worker maker. Experimental results show that the algorithm is efficient and scalable on large-scale computations.