Application of High Performance Computation for the Prediction of Urban Area Earthquake Disaster

With the aim of simulating earthquake disaster of a vast urban area, we enhanced the Integrated Earthquake Simulator (IES) with high performance computation (HPC). IES is a system to seamlessly simulate possible earthquake hazard, disaster, evacuation and recovery. The need of such system is indispensable. Businesses, local and state governments need reliable quantitative predictions of infrastructural damages due future earthquakes, to minimize damages and have sound recovery plans to minimize socioeconomic aftermaths. Only a system like IES can make reliable predictions taking the latest information and the current built-in-environment to the account; conventional predictions cannot take the current built-in-environment to the account. In order to simulate a series of scenario earthquakes within a reasonable short time, which is necessary for reliable quantitative predictions, IES should be enhanced to utilize HPC resources. Though IES has been enhanced with standard parallel computing techniques, some bottlenecks seriously limited it's scalability to double digit numbers of CPUs. We significantly improved its parallel computing performance by eradicating all the major bottlenecks. Details of the bottlenecks, remedies implemented in the modified IES and other performance enhancements are presented. Scalability of the modified IES is demonstrated with representative GIS tiles from Shinjuku district of Tokyo. It is shown that parallel computing performance of IES is improved significantly both in terms of problem size and run time. According to rough estimations, simulation of 2-3 million buildings in Tokyo needs ten to fifteen thousands of CPUs.