Parallel data transposition in numerical algorithm for solving the Gross-Pitaevski equation

Abstract

<p>Ova doktorska teza se bavi proučavanjem i razvojem paralelnih<br />algoritama za transponovanje distribuiranih trodimenzionalnih<br />struktura podataka, kao i implementacijom ovih algoritama u okviru<br />C/OpenMP/MPI programske paradigme. Razvijena implementacija je<br />primenjena na rešavanje nelinearne parcijalne diferencijealne<br />jednačine Šredingerovog tipa (Gros-Pitaevski jednačina) korišćenjem<br />Krenk-Nikolson metoda, a u okviru teze je predstavljen ciklus razvoja<br />odgovarajućeg softvera, kao i rezultati testova validnosti i merenja<br />performansi dobijenih na računarskom klasteru.</p>

<p>This thesis studies and develops parallel algorithms for transposing<br />distributed three-dimensional data structures, and describes their technical<br />implementation in C/OpenMP/MPI programing paradigm. The developed<br />implementation is applied for solving of nonlinear partial differential equation<br />of the Schroedinger type (Gross-Pitaevskii equation) using Crank-Nicolson<br />method. The thesis presents the corresponding software development cycle,<br />as well as results of validity tests and performance measurements obtained<br />on a computer cluster.</p>