Parallel scalability enhancements of seismic response and evacuation simulations of integrated earthquake simulator

Abstract

We developed scalable parallel computing extensions for Seismic Response Analysis (SRA) and evacuation simulation modules of an Integrated Earthquake Simulator (IES), with the aim of simulating earthquake disaster in large urban areas. For the SRA module, near ideal scalability is attained by introducing a static load balancer which is based on the previous run time data. The use of SystemV IPC as a means of reusing legacy seismic response analysis codes and its impacts on the parallel scalability are investigated. For parallelizing the multi agent based evacuation module, a number of strategies like communication hiding, minimizing the amount of data exchanged, virtual CPU topologies, repartitioning, etc. are used. Priliminary tests on the K computer produced above 94% strong scalability, with several million agents and several thousand CPU cores. Details of the parallel computing strategies used in these two modules and their effectiveness are presented. © 2013 Springer-Verlag.