Fig. 3
From: Parallelisation strategies for agent based simulation of immune systems
FLAME GPU state diagram for the Monte Carlo implementation of pairwise cell interaction model. The diagram shows the order of agent functions (black rectangles) and interactions between agents via message lists (coloured rhombuses) through a single iteration of the main simulation loop. Within a layer, indicated by blue dashed boxes, functions may execute concurrently. For the Monte Carlo implementation, A agents and B agents both execute their respective Generate Rank functions in the first layer. The ranks of each respective agent type are output in to respective message list, A Rank and B Rank. In the second layer, agents from each population iterate the message list from their own population to find the location within the rank-list, before broadcasting this information to the other type of agent via the relevant Find Position + Broadcast function. Finally, in the third layer, Agents iterate messages form the opposing population to determine if they are able to interact based on their position within their population and also information about the opposing population