IN43B-1728
Massive Cellular Automata in Geosimulation: Antarctica Ice Melting as Example
Thursday, 17 December 2015
Poster Hall (Moscone South)
Hai Lan, University of Maryland College Park, College Park, MD, United States
Abstract:
One of the essential features of the cellular automata (CA) model is its high scalability: CA lattices can be theoretically run at gargantuan size to represent intricacies of complex phenomena. However, one barrier in the use of cellular automata for scientific simulations is the issue of scalability in terms of the number of cells, to either model phenomena at finer granularities or at larger scales. Some researchers have developed parallel CA algorithms using MapReduce to eke out efficiency, but MapReduce may not provide the ideal scheme to address messy parallelism in large CA when they require complex rule-sets and broker a lot of state exchange across large solution-space lattices. In this research, we take advantage of the Bulk Synchronous Parallel (BSP) model of distributed computation, via the Giraph open-source implementation, to implement large-scale cellular automata simulations. Additionally, this study also describes a scientifically interesting example, in which ice dynamics in Antarctic is simulated using a melting model. Short-term and medium-term ice sheet dynamics are driven by a variety of forces. We do not fully understand what they might be and how they interplay, and simulation is an important medium for building the science to guide us in finding answers. In our experiments, using a voxel CA comprising 1 trillion cells—by far the largest scale voxel-based CA model reported in literature—which took only 2.48 minutes for per step for processing.