Distribution-invariance in agent-based spatial simulations

Large-scale spatial simulations of social beings often exceed the processing capabilities of a single computer. Efficient parallelization and distribution of such simulations is typically based on model partitioning and parallel processing of model parts on many computing nodes. In simulation models which consider mutually exclusive actions of agents, such an approach raises significant problems related to synchronization of the borders between environment partitions processed by different computing nodes. This paper addresses this problem and presents an abstract method for achieving distribution transparency in discrete spatial agent-based simulations. It proposes definitions of distribution-invariance in such simulations and a distribution-invariant simulation algorithm together with formal frames to express the model compliant with the algorithm. A verification of the method in the context of the proposed definitions is presented as well as an analysis of the scalability and applicability of the method.