Towards Fine-grained Spatial Partition for Wildfire Simulation

Tasks partitioning is one of the most challenging issues in large scale parallel simulations. The simulation performance may vary dramatically if different partition schemes are used. This paper presents a spatial partition method named fine-grained spatial partition for parallel simulation of large scale wildfire. The fine-grained spatial partition is inspired by the four color theorem. It divides a cell space into many smaller regions and allocates adjacent regions to different processing units. Because no adjacent regions are on the same processing units, a fire started in one region will quickly spreads out to adjacent regions and thus achieves a higher degree of parallelism regardless where the fire is started. Experiment results are collected and compared with that of a uniform spatial partition.