A New Opportunity to Urban Evacuation Analysis: Very Large Scale Simulations of Social Agent Systems in Repast HPC

Understanding the dynamics of urban evacuation systems -due to catastrophic disasters induced by forces of nature like flooding or tsunamis, terrorism or nuclear power plant accidents- has elicited massive interest over the past years. While discrete event simulations of evacuation models become prohibitively complex dealing with the time, space and individual behavior, multiagent based models have revealed to be a potentially more effective. This paper introduces models of configurations of social agents at a massive scale, which, together with the most recent supercomputing technology allows for a simulation analysis of realistic evacuation models at the level of large cities (10^6-10^8). Agent based models of demographics and the morphology of cities together with population densities, mobility patterns, individual decision making and agent interactions are implemented into a tool chain which ultimately generates Repast HPC code, which is then executed on 2048 node shared memory multiprocessor (SGI Altix Ultra Violet 1000). We demonstrate how different evacuation strategies can be assessed based on costly, yet feasible simulation runs - thus evidencing, that a whole class of demanding, very complex simulation problems has found a convincing solution.