Parallelization of lattice Boltzmann software for execution on multi-GPU clusters with application to the simulation of blood flow through human arteries

Abstract

It is important to consider the blood flow pattern when planning vascular interventions of atherosclerotic plaques. Large scale computer modeling can be very helpful in this case. The software presented in this paper numerically models blood flow through patient-specific blood vessels. Lattice Boltzmann method was used to simulate blood flow. The principles of GPU (Graphics Processing Unit) programming are applied during implementation and the developed software was parallelized using the CUDA (Compute Unified Device Architecture) and optimized to run on a multi-GPU cluster using the MPI approach. Using the multi-GPU infrastructure, numerical simulations can utilize larger amount of memory resources for the computation, making the level of reality of the models an order of magnitude higher. Execution of the presented software enables fast and reliable case-studies and parametric analyses useful for medical decision-making. The presented software can give medical professionals fast quantitative information about fluid flow in diseased arteries and assist them in selecting the most appropriate treatment.