Parallelism management in cellular automata networks

In this paper we present some preliminary results obtained when dealing with simulation applications that use the Cellular Automata (CA) paradigm. In particular we used the Cellular Automata Network (CAN) model, an extended CA model, that allows to model complex physical systems that can be represented in terms of connected components. In CAN model each component is represented by a cellular automaton, while interactions among components are represented by a network of cellular automata. Simulation applications written according to the CAN model offer potentially two kinds of parallelism: one is the data parallelism intrinsic to the standard CA model, the other is the control parallelism coming from the possibility to concurrently execute more automata of the network under some conditions that will be explained in the paper. In order to obtain better performances of CAN applications running on a target parallel machine with a fixed amount of computational resources, a mapping between the potential parallelism and the available resources is necessary. We show how to manage the two kinds of parallelism in a real CAN application, simulating a phenomenon of colloidal aggregation, to improve application performances.