Parallel CPU-GPU Implementation of a Model for Whole Blood

This project is about the computational aspects of a model for whole
blood. Blood is a suspension consisting of a fluid (plasma) and a
solid phase suspended in it (red blood cells, platelets, monocytes).
In an existing 3D blood model in our group the fluid phase is modeled
using a Lattice-Boltzmann (LBM) implementation which is tighly coupled
to an FEM model for the deformable cells. Due to its regular lattice
and local interactions, LBM it is very well suited for a
parallelization on GPU’s, whereas the computational structure of the
FEM model might still demand a CPU implementation. One of the
challenges of coupling both is the efficient exchange of information
between the two models running on different hardware. Another
challenge is the parallelization of these models. For the fixed-grid
LBM, boundary information has to be exchanged, whereas in the cell
model particles will have to be exchanged between the nodes of the
cluster. Several strategies to achieve this goal are possible
including multiscale modeling approaches. GPU implementations of LBM
exist that could be used (such as Sailfish). The CPU-GPU blood model
implementation will then be run on a GPU cluster for HPC simulation
studies of blood rheology and transport properties in blood.