In a large-scale pool fire simulation, the processes that must be modeled are complex and coupled. The flow is often highly turbulent, dynamic vortical structures are present, the chemical reactions involve several thousand elementary steps and hundreds of species/intermediates, and radiation, the dominant mode of heat transfer, is strongly affected by the presence of soot. The range of length and time scales associated with all these processes cannot be simulated on even the most powerful supercomputers available today. Our approach to making this intractable problem tractable has been twofold: one, to improve the models used at all levels in the simulation (i.e., transport models and subgrid scale models) and two, to parallelize the simulation tool to run on massively parallel machines. We have employed Large Eddy Simulation (LES) to model the fluid dynamics and the convection-diffusion scalar transport. LES successfully captures the transient nature of the coherent vortical structures present in a pool fire. We have integrated these improved models into a computational framework that provides support for parallelization. Preliminary validation results show the capability of the fire simulation tool to capture the puffing nature of pool fires. In addition, scalability studies of the simulation tool reveal close to linear scalability up to 500 processors.