Analysis of the impact of using high performance computing in fire modeling

Abstract

The modeling of fire development and smoke spreading in large buildings, with high geometric complexity, using CFD specific methods, represents a computationally very demanding task. In most real fire scenarios, problems associated to FDS model require the division of complex geometry into compartments or zones, each being associated with one or more discretization network. These multiple mesh models generate long execution times of days or even weeks. In most cases, time is a very precious parameter in fire safety design studies or postevent technical expertise, engineers having to deliver results with maximum accuracy in a shortest time possible. These performance gains can only be achieved using High Performance Computing clusters and robust computing infrastructures, including advanced, very low latency networking and data storage systems. This paper aims to highlight the benefits of parallel processing (mainly of distributed processing) in the case of fire modeling using Fire Dynamic Simulator (FDS), developed by National Institute of Standards and Technology. To quantify the contribution of such information infrastructures, the study used a fire scenario occurred in a building with complex geometry, virtual model on which were applied optimization techniques of computational meshes for multi node cluster server configurations, or partitioning (assigning the appropriate number of meshes or volumes) between each cluster nodes (MPI process). Performance gains were characterized through the three specific parameters mentioned in the literature: speedup ratio (Sn), parallel efficiency (En) and scalability.