A Parallel, Distributed, High-Performance Architecture for Simulating Particle-Based Models

摘要

Particle-based models are widespread in the field of computer graphics and are mostly used in soft-body dynamics, for simulating surfaces such as cloth, fluids and biologic tissue. As model resolution and scenario complexity increases, the computation required for these particular applications becomes overwhelming for a single processing unit, especially when interactivity is required, thus parallelization must be employed in order to provide a fast, flexible and scalable simulation environment. High-performance computing architectures such as graphics clusters may provide the parallel computing and rendering power required, but the distributed and remote nature of the computation and rendering process introduce specific challenges that must be tackled. We propose a parallel, distributed, modular system architecture for a particle-based simulator on GPU clusters, encapsulating powerful parallel and distributed processing, distributed rendering and remote interaction techniques, for flexible, fast simulation of large models and complex scenarios. For validating and evaluating the proposed architecture, we perform a visual comparison of two largely used numeric integration methods, namely the explicit Velocity Verlet and implicit Euler integration techniques.