This thesis presents interdisciplinary research carried out in the areas of physics and computer science. We introduce a new model based on cellular automata (CA) to describe laser dynamics, which is an alternative to the standard description based on differential equations. We also study how to take advantage of the intrinsic parallel nature of CA to carry out high perfor- mance computational simulations with that model on parallel and distributed computers.
CA are a class of fully discrete, spatially-distributed dynamical systems charac- terized by local interaction and synchronous parallel dynamical evolution. Our CA-based model reproduces the main laser dynamics phenomenology. It reinforces the vision of laser as a complex system, since the macroscopic properties of the laser system emerge with this model as a cooperative phe- nomenon from elementary components that interact locally under simple rules. This model represents a new methodological approach for laser modelling, which can be advantageous in cases in which the differential equations are difficult or impossible to integrate, or not entirely applicable because the usual approximations are not valid. It also opens the door to new more detailed CA-based models that could capture more features of laser behaviour.
We investigate how to parallelize the model efficiently. A parallel implementa- tion has been developed using the message passing paradigm. Its performance and scalability have been analysed for the two most usual kinds of platforms for this kind of applications: dedicated clusters, and heterogeneous non-dedicated clusters including a dynamic load balancing strategy. In the latter case a very flexible modular approach has been employed in which the model is executed on top of a dynamic load balancing tool. The results confirm that, in spite of potential difficulties, it is feasible to execute large, fine-grained simulations with CA-based models—including the laser model described—on the most usual types of cluster computing platforms with a good efficiency and scalability.