Parallel computation for tsunami

Abstract

In a simulation of wave propagation in a large fluid domain like the case of tsunami, the implementation of wave model for such large computational domain has been a challenging task. Because of the domain size with several unknown variables, wave simulation requires large amounts of computational time and memory. In such case a single computer with limited memory size is inadequate to carry out the computations. To make such computational demand possible, parallel code is developed with the use of domain decomposition technique and the message passing interface (MPI). The Boussinesq wave model for the numerical simulation of tsunami propagation is implemented based on a finite difference scheme. Parallel tridiagonal solver and efficient data transfer scheme are used to improve the performance of solving tridiagonal systems. Two tsunami events, the 1999 Vanuatu tsunami and the 2004 Indian Ocean tsunami are used to test the parallelized program. Experimental results indicate that parallel code running on more than one processor gives much better performance and is found to fit the tsunami model very satisfactorily in comparison with running on a single processor.