Scalable HPC Enhanced Agent Based System for Simulating Mixed Mode Evacuation of Large Urban Areas

Abstract

Abstract This short paper presents an HPC enhanced Agent Based Model (ABM) developed with the aim of quantitatively estimating the strategies for accelerating emergency mass evacuations, like tsunami evacuation. In order to facilitate inclusion of various influencing factors, such as localized congestion, multi-modes, pedestrian vehicle interactions, fallen debris from damaged buildings, visibility, etc., which demand detailed models, the developed system includes a 1mx1m resolution model of environment, and agents capable of perceiving and autonomously interact with this high resolution environment and visible agents. In order to meet the computational demand of large scale simulations with complex agents, a scalable high performance extension was implemented. Short introductions to the agent based model and the HPC extension are presented in this paper. In order to demonstrate the scalability of the system, both in problem size and computational capability, a 588 km2 region with 10 million agents is simulated in K computer. It is demonstrated that the system has high strong scalability up to 2048 computing nodes, which is equivalent to 16,384 CPU cores.