Load Balancing in Parallel Implementation of Hydraulic Erosion Visual Simulations

A load balancing in a parallel implementation of an algorithm visually simulating hydraulic erosion is introduced. Water flowing over the terrain surface captures particles of material and changes their location that has a great influence to the terrain morphology. The algorithms attempting to solve this processes require a high computational effort. A data parallel approach is used. Data representing the terrain is divided into strips and the work is assigned to different computational units. The work is asymmetrical, the parts of the terrain keeping water require higher computational effort than the dry ones. The chunks of data are evaluated and assigned to each computational unit in such a way to maintain equilibrium of the work distribution. At the end of every parallel step the data is collected, load balancing evaluated, and the work is redistributed again. The computational units communicate by message passing. The surface of Mars’s data obtained from NASA Mars Orbital Laser Altimeter is used and we simulate water running over the terrain. We run the implementation on IBM R6000 with 12 CPUs and the runtime of the computational units with the load balancing differs approximately 10 %.