Караваев Д.А.   Kulikov I.M.   Yakimenko A.A.  

Scalable Parallel Implementation of 3D Seismic Wave Propagation Simulation Algorithm on Hybrid Cluster.

Reporter: Караваев Д.А.

Scalable Parallel Implementation of 3D Seismic Wave Propagation Simulation Algorithm on Hybrid Cluster.
D. A. Karavaev1, I. M. Kulikov1,2, A. A. Yakimenko1,2
1ICMMG SB RAS, Novosibirsk
2Novosibirsk State Technical University, Novosibirsk
kda@opg.sscc.ru, kulikov@ssd.sscc.ru, al--le@yandex.ru

In the paper, we consider a scalable parallel realization of a numerical method for forward 3D seismic simulation.  We solve the system of equations of dynamic theory of elasticity, which is presented in terms of velocities of displacement and strain tensor components. We apply 3D domain decomposition and a difference method [1]. We have developed a high-scalable parallel algorithm to use hybrid clusters with graphical processors, GPUs. Such a realization use MPI and CUDA to perform computations on GPUs. We present results of studying the parallel algorithm implementation on the GPU cluster of the SSCC SB RAS. The developed technology of large-scale simulation for 3D numerical modeling including mathematical method and efficient parallel realization can be applied to perform computation on other types of computing devices of a hybrid clusters. The acceleration achieved is a promising approach to solve large problems and use in computations hundreds and thousands of cores of HPC clusters [2]. Calculations were carried out on the cluster of SSCC SB RAS.
The work was partially supported by RFBR grants No. 14-07-00832, 14-05-00867, 15-07-06821, 15-31-20150 and MON RK/GF4.

1. Virieux J. P-SV wave propagation in heterogeneous media: Velocity-stress finite- difference method // Geophysics. Volume 51.Number 4. 1986. pp. 889-901.
2. Chernykh I., Glinskiy B., Kulikov I., Marchenko M., Rodionov A., Podkorytov D., Karavaev D.: Using Simulation System AGNES for Modeling Execution of Parallel Algorithms on Supercomputers// Computers, Automatic Control, Signal Processing and Systems Science. The 2014 Int. Conf. on Applied Mathematics and Computational Methods in Engineering. pp. 66-70.

To reports list