On-the-fly clustering for exascale molecular dynamics simulations.

Hello everyone. I am Cheria, and I oversee the computational aspects of our projects, particularly in simulation tools. Our journey began with the development of our Molecular Dynamics code, ExaMD, which emphasizes high-performance computing capabilities. Over the years, we have created numerous massively parallel algorithms and GPU ports for most of these algorithms. This advancement has sparked interest in optimizing other simulation tools to leverage similar parallel algorithms tailored for specific simulations.

Currently, we have established a specialized N-body platform called ExaNBody. This platform serves as a foundation for any N-body simulation code. To date, we have developed three simulation codes based on ExaNBody, which can simulate various systems, including small grains of matter, rigid bodies such as sand or powders, and more complex interactions involving rigid bodies with contact dynamics, friction, and collision detection. In addition to our Molecular Dynamics code, we are also exploring hydrodynamics through particle simulation techniques, specifically smooth particle hydrodynamics, which address the limitations of mesh-based methods.

We have created a robust ecosystem for developing our simulation software tailored for high-performance computing. This ecosystem consists of three main components. First, we utilize industry-standard low-level libraries, including MPI, OpenMP, C++, YAML for user inputs, and CUDA and HIP for GPU acceleration. Built on top of these low-level standards, we have developed three software layers. The first is OnikaBrick, which serves as the interface with the hardware and incorporates a component-based programming model. The second layer is ExaNBody, which focuses on N-body parallel algorithms, and on top of this, we have our three simulation code applications.