ABSTRACT

This research investigates computational performance optimisation of the Fire Dynamics Simulator (FDS) for full-scale compartment fire analysis through parallel processing. By employing high-performance computing (HPC) resources and systematic mesh decomposition strategies, the study evaluates how different meshing resolutions and processor allocations affect simulation speed and accuracy. A full-scale compartment fire scenario, validated against experimental data, serves as the reference case. Simulations are conducted on the Rangpur HPC platform using varying core counts (1-64) and mesh resolutions (5 cm, 10 cm, 20 cm). Speedup, communication overhead, load balancing, and decomposition strategy are quantitatively analysed. Results show that finer mesh resolutions benefit more from parallelism due to higher computational load per mesh, whereas coarse meshes suffer from disproportionate communication times, limiting parallel efficiency gains. Load balancing is assessed using Load Imbalance Factor (LIF), indicating that meshing strategy significantly affects idle times across processors. Simulation segments near fire sources or airflow interfaces consistently demand greater computational resources, underscoring the importance of strategic mesh zoning. The study concludes with practical recommendations for mesh partitioning and processor allocation, aiming to reduce run times without compromising simulation fidelity. This guidance enhances the accessibility of detailed fire modelling in complex built environments, improving the integration of FDS in fire safety engineering workflows.