Fire Dynamics Simulation is one of the various CFD application based on solving Navier-Stokes equation. It specifically focuses on low-speed and thermally driven flow with intent to describe smoke and heat transport from fires.
There are multiple software stack available on the market running on both Windows and Linux to create geometries, run the simulation and visualize the results.
Nevertheless, the simulation itself, for the majority of these software, is based on the Open source Fire Dynamics Simulator (FDS) and Smokeview (SMV) from the National Institute of Standards and Technology (NIST)
The simulation runtime on a single workstation can vary from days to months on a real scenario and is a great fit for a High Performance Computing (HPC) approach since the simulation can take advantage of distributed (Parallel) computing.
AWS provides a wide range of HPC services such as AWS ParallelCluster 2.7.0, Amazon FSx for Lustre, Elastic Fabric Adapter (EFA) and NICE DCV to provide scaling, performance and graphics-accelerated environment for heavy CFD workloads.
In this technical post, I walk through these services, providing scripts and templates to set up a HPC system with EFA and Amazon FSx for Lustre to run a Fire Dynamics Simulation. I will also detail some best practices to optimize scalability and performance of the simulation itself.