High-Performance Parallel Framework for Simulating Deformable Membranes in Fluid Flow

Parallel block 1, Session A1: 11:00 am

P. Bhagawat, D. Tafti  
Presenter: Bhagawat, Pooja

Simulating highly deformable membranes in fluid flow presents significant computational challenges due to complex fluid-structure interactions. When performed in distributed-memory parallel computing environments, these simulations introduce additional challenges associated with inter-processor communication overhead. This work presents the development of a high-performance parallel computational framework designed to improve the efficiency and scalability of such simulations, using red blood cells (RBCs) as a representative application. The framework combines a finite volume method for fluid flow, a coarse-grained membrane model, and a direct-forcing immersed boundary method for fluid–structure coupling. Two computational strategies are developed to improve parallel performance: a stage-specific dynamic ghost extent strategy that minimizes inter-processor communication based on spatial data dependencies, and a topology-based neighbor-search algorithm that reduces the computational cost of intercellular interactions. The framework is validated against established RBC benchmarks and evaluated across different hematocrit levels. The proposed strategies reduce computational and communication overhead, achieving approximately 76% parallel efficiency on 1000 CPU cores. These developments provide a scalable computational approach for simulating deformable membranes in fluid flow.