(RP24) Approximating a Stochastic Cellular Automaton Using Parallel Discrete Event Simulation
TimeTuesday, June 18th8:30am - 10am
DescriptionStochastic Cellular Automaton (SCA) represents an important class of simulations that is used to model the probabilistic state change of dynamic systems such as sand dune formation, fire, snow, and wind propagation. In SCA, the state of the collection of entities is updated based on a stochastic rule, which means the new entities’ states are chosen according to some probability distributions. ReSCAL (Real-Space Cellular Automation Laboratory) is a sequential cellular automaton modeling tool in real 3d space, primarily designed for geomorphology such as how sand dunes form when wind blows. The SCA modeling tool has been designed to be sufficiently generic in its approach to deal with a wide range of applications. This research poster is a work-in-progress that aims at parallelizing the ReSCAL simulation framework to leverage on-node scaling on several different architectures. The challenge in developing a parallel framework for SCA is that since it simulates asynchronous first-neighbor interactions based on Markov chains, several rules for next-state transition can be conflicting. This study investigates heuristics to parallelize the ReSCAL framework, and shows that on-node scaling on IBM power 9 and Intel Sandy Bridge architectures can reasonably improve the speed of the framework. These preliminary results serve as a motivation for porting the code to GPUs and developing strategies for parallelizing probabilistic simulations further.