The overlapping algorithm for solving shallow water equations on massively-parallel architectures with distributed memory

Abstract

The paper proposes and tests algorithm for acceleration shallow water equations on parallel computers with distributed memory. The algorithm is based on the optimization of computations, interprocessor communications and the amount of transferred information, which can significantly improve the scalability of computer code. The tests show that the overlapping algorithm provides serious advantages for large communicator sizes of the model when the acceleration strongly depends of communications delays. Algorithm is realized as one of the blocks of parallel eddy-resolving World Ocean Model. The paper was recommended for publication by the program committee of the International Supercomputer Conference "Scientific service on the Internet: exafplops future."